1
00:00:05,415 --> 00:00:06,917
GOOD AFTERNOON.  

2
00:00:06,917 --> 00:00:07,551
I'M CATHERINE LAW, DIRECTOR OF 

3
00:00:07,551 --> 00:00:08,185
COMMUNICATIONS AT THE NATIONAL 

4
00:00:08,185 --> 00:00:08,785
CENTER FOR COMPLEMENTARY AND 

5
00:00:08,785 --> 00:00:10,621
INTEGRATIVE HEALTH, NCCIH, AT 

6
00:00:10,621 --> 00:00:11,188
THE NATIONAL INSTITUTES OF 

7
00:00:11,188 --> 00:00:18,511
HEALTH.

8
00:00:18,511 --> 00:00:19,312
I WANT TO WELCOME YOU TO THE 

9
00:00:19,312 --> 00:00:20,880
LATEST PRESENTATION IN NCCIH'S 

10
00:00:20,880 --> 00:00:21,948
INTEGRATIVE MEDICINE RESEARCH 

11
00:00:21,948 --> 00:00:27,776
LECTURE SERIES.

12
00:00:27,776 --> 00:00:28,410
TODAY'S LECTURE, "HARNESSING AI 

13
00:00:28,410 --> 00:00:33,292
TO EXPLORE HEALTH RESTORATION IN

14
00:00:33,292 --> 00:00:34,360
DIABETES" IS PRESENTED BY DR. 

15
00:00:34,360 --> 00:00:35,428
AARON LEE FROM THE UNIVERSITY OF

16
00:00:35,428 --> 00:00:39,603
WASHINGTON.

17
00:00:39,603 --> 00:00:41,172
BEFORE WE INTRODUCE OUR SPEAKER,

18
00:00:41,172 --> 00:00:45,910
A FEW BRIEF REMINDERS ABOUT 

19
00:00:45,910 --> 00:00:46,444
LOGISTICS FOR TODAY'S 

20
00:00:46,444 --> 00:00:47,479
PRESENTATION.

21
00:00:47,479 --> 00:00:50,529
FIRST, A REMINDER THAT THIS 

22
00:00:50,529 --> 00:00:51,530
PRESENTATION IS BEING RECORDED 

23
00:00:51,530 --> 00:00:52,598
AND WILL BE ARCHIVED ON THE NIH 

24
00:00:52,598 --> 00:00:54,200
VIDEOCAST WEBSITE FOR FUTURE 

25
00:00:54,200 --> 00:01:01,755
VIEWING.

26
00:01:01,755 --> 00:01:03,090
IF YOU HAVE QUESTIONS TODAY, YOU

27
00:01:03,090 --> 00:01:03,724
ARE WELCOME TO SUBMIT A QUESTION

28
00:01:03,724 --> 00:01:04,691
AT ANY TIME USING THE VIDEOCAST 

29
00:01:04,691 --> 00:01:09,932
FEEDBACK FORM.

30
00:01:09,932 --> 00:01:10,566
THE LINK SHOULD BE LOCATED JUST 

31
00:01:10,566 --> 00:01:11,200
BELOW YOUR SCREEN FOR VIEWING 

32
00:01:11,200 --> 00:01:15,328
THE LIVE STREAM.

33
00:01:15,328 --> 00:01:15,962
YOU MAY ALSO SEND QUESTIONS VIA 

34
00:01:15,962 --> 00:01:21,409
E-MAIL TO 

35
00:01:21,409 --> 00:01:27,108
NCCIHEVENTS@MAIL.NIH.GOV.

36
00:01:27,108 --> 00:01:28,309
WE WILL TRY TO ANSWER AS MANY 

37
00:01:28,309 --> 00:01:28,943
QUESTIONS FROM THE AUDIENCE AS 

38
00:01:28,943 --> 00:01:32,878
WE CAN.

39
00:01:32,878 --> 00:01:34,112
IF WE MISS ANY QUESTIONS, WE 

40
00:01:34,112 --> 00:01:34,713
WILL SHARE THEM WITH DR. LEE 

41
00:01:34,713 --> 00:01:41,631
AFTER THE EVENT.

42
00:01:41,631 --> 00:01:46,381
WITH THOSE REMINDERS COVERED, I 

43
00:01:46,381 --> 00:01:46,981
AM NOW PLEASED TO INVITE DR. 

44
00:01:46,981 --> 00:01:47,615
DAVID SHURTLEFF, DEPUTY DIRECTOR

45
00:01:47,615 --> 00:01:48,483
ON NCCIH, TO INTRODUCE OUR 

46
00:01:48,483 --> 00:01:54,124
SPEAKER.

47
00:01:54,124 --> 00:01:57,227
DR. SHURTLEFF.

48
00:01:57,227 --> 00:01:58,829
>> THANK YOU, CATHERINE.

49
00:01:58,829 --> 00:02:00,797
WELCOME, EVERYONE.

50
00:02:00,797 --> 00:02:01,431
MY NAME IS DAVID SHURTLEFF AND 

51
00:02:01,431 --> 00:02:07,137
I'M THE DEPUTY DIRECTOR OF 

52
00:02:07,137 --> 00:02:07,337
NCCIH.

53
00:02:07,337 --> 00:02:07,904
WELCOME TO OUR SUMMER 2024 

54
00:02:07,904 --> 00:02:09,272
LECTURE IN NCCIH'S INTEGRATIVE 

55
00:02:09,272 --> 00:02:12,476
MEDICINE RESEARCH LECTURE 

56
00:02:12,476 --> 00:02:15,145
SERIES OR IMLS.

57
00:02:15,145 --> 00:02:15,779
I'M HAPPY TO TELL YOU THAT THIS 

58
00:02:15,779 --> 00:02:16,880
YEAR MARKS 25 YEARS SINCE THE 

59
00:02:16,880 --> 00:02:17,781
FOUNDING OF NCCIH, AND TODAY'S 

60
00:02:17,781 --> 00:02:20,450
LECTURE IS AN EXAMPLE OF THE 

61
00:02:20,450 --> 00:02:21,017
POTENTIAL IMPACT OF A WHOLE 

62
00:02:21,017 --> 00:02:21,651
PERSON APPROACH TO RESEARCH AND 

63
00:02:21,651 --> 00:02:32,195
THE POTENTIAL IMPACT ON HEALTH.

64
00:02:36,299 --> 00:02:42,205
DR. LEE'S TOPIC TODAY IS

65
00:02:42,205 --> 00:02:42,839
DR. LEE'S TOPIC REFLECTS SEVERAL

66
00:02:42,839 --> 00:02:43,440
AREAS OF RESEARCH INTEREST AT 

67
00:02:43,440 --> 00:02:43,774
NCCIH.

68
00:02:43,774 --> 00:02:44,841
THE FIRST IS SALUTOGENESIS, OR 

69
00:02:44,841 --> 00:02:47,144
RESTORATION OF HEALTH.

70
00:02:47,144 --> 00:02:49,780
AS MANY OF YOU MAY KNOW, AT 

71
00:02:49,780 --> 00:02:51,915
NCCIH, WE LOOK AT

72
00:02:51,915 --> 00:02:52,883
MULTIPLE FACTORS THAT PROMOTE 

73
00:02:52,883 --> 00:02:56,686
EITHER HEALTH OR DISEASE AND 

74
00:02:56,686 --> 00:02:57,254
SCIENTIFICALLY CONSIDER THE 

75
00:02:57,254 --> 00:02:59,923
WHOLE PERSON AS A COMPLEX SYSTEM

76
00:02:59,923 --> 00:03:02,325
THAT CAN EXIST IN A CONTINUUM OF

77
00:03:02,325 --> 00:03:04,127
HEALTH AND DISEASE WITH THE GOAL

78
00:03:04,127 --> 00:03:09,065
OF RESTORING HEALTH OF ALL 

79
00:03:09,065 --> 00:03:11,535
PEOPLE.

80
00:03:11,535 --> 00:03:19,342
IT IS GREATLY APPRECIATED.

81
00:03:19,342 --> 00:03:21,845
THROUGH PATHOGENESIS, WE KNOW 

82
00:03:21,845 --> 00:03:25,081
ABOUT DISEASES ABOUT ORGANS.

83
00:03:25,081 --> 00:03:26,216
SALUTOGENESIS FOCUSES ON THE 

84
00:03:26,216 --> 00:03:31,054
ORIGINS OF HEALTH, RATHER THAN 

85
00:03:31,054 --> 00:03:31,288
DISEASE.

86
00:03:31,288 --> 00:03:33,723
AND FACTORS THAT SUPPORT THE 

87
00:03:33,723 --> 00:03:37,060
HUMAN BEING WHICH IS LESS 

88
00:03:37,060 --> 00:03:42,032
STUDIED AND LESS UNDER.

89
00:03:42,032 --> 00:03:43,567
WHOLE PERSON HEALTH RESEARCH 

90
00:03:43,567 --> 00:03:45,869
INVOLVES THE STUDY OF THE WHOLE 

91
00:03:45,869 --> 00:03:48,572
PERSON, NOT SEPARATE ORGANS AND 

92
00:03:48,572 --> 00:03:49,973
CONSIDERS MULTIPLE SYSTEMS THAT 

93
00:03:49,973 --> 00:03:52,242
CAN PROMOTE HEALTH OR DISEASE, 

94
00:03:52,242 --> 00:03:54,177
WHICH LEADS TO RESEARCH ON 

95
00:03:54,177 --> 00:03:55,278
MULTIMODAL APPROACHES IN 

96
00:03:55,278 --> 00:03:56,880
DIFFERENT SYSTEMS AND 

97
00:03:56,880 --> 00:03:58,949
DISCIPLINES, PROMOTING HEALTH AS

98
00:03:58,949 --> 00:04:01,084
A WHOLE BODY PHENOMENON.

99
00:04:01,084 --> 00:04:07,858
FINALLY, NCCIH IS INTERESTED IN 

100
00:04:07,858 --> 00:04:10,427
APPLYING AND USING ARTIFICIAL 

101
00:04:10,427 --> 00:04:11,461
INTELLIGENCE, MACHINE LEARNING, 

102
00:04:11,461 --> 00:04:14,231
AND USE OF BIG DATA TO ADDRESS 

103
00:04:14,231 --> 00:04:15,599
THE CHALLENGES OF PROMOTING 

104
00:04:15,599 --> 00:04:18,034
HUMAN HEALTH, WHICH WE WILL HEAR

105
00:04:18,034 --> 00:04:24,875
MORE ABOUT FROM DR. LEE IN JUST 

106
00:04:24,875 --> 00:04:31,948
A MOMENT.

107
00:04:31,948 --> 00:04:32,549
AND NOW IT IS MY PLEASURE TO 

108
00:04:32,549 --> 00:04:35,886
INTRODUCE DR. AARON LEE.

109
00:04:35,886 --> 00:04:37,120
WITHIN THE DEPARTMENT OF 

110
00:04:37,120 --> 00:04:39,522
OPHTHALMOLOGY

111
00:04:39,522 --> 00:04:40,156
AT THE UNIVERSITY OF WASHINGTON 

112
00:04:40,156 --> 00:04:40,690
IN SEATTLE, DR. LEE IS AN 

113
00:04:40,690 --> 00:04:43,226
ASSOCIATE PROFESSOR, THE C. DAN 

114
00:04:43,226 --> 00:04:43,760
AND IRENE HUNTER ENDOWED 

115
00:04:43,760 --> 00:04:45,061
PROFESSOR, AND A VITREORETINAL 

116
00:04:45,061 --> 00:04:46,930
SURGEON.

117
00:04:46,930 --> 00:04:47,564
TRANSLATION OF NOVEL COMPUTATION

118
00:04:47,564 --> 00:04:48,198
TECHNIQUES IN MACHINE LEARNING 

119
00:04:48,198 --> 00:04:49,065
TO UNCOVER NEW DISEASE 

120
00:04:49,065 --> 00:04:50,634
ASSOCIATIONS AND MECHANISMS FROM

121
00:04:50,634 --> 00:04:51,101
ROUTINE CLINICAL DATA.

122
00:04:51,101 --> 00:04:51,668
EXAMPLES INCLUDE ELECTRONIC 

123
00:04:51,668 --> 00:04:52,235
HEALTH RECORDS AND IMAGES.

124
00:04:52,235 --> 00:04:53,303
DR. LEE RECEIVED HIS BACHELOR'S 

125
00:04:53,303 --> 00:04:54,337
DEGREE FROM HARVARD UNIVERSITY 

126
00:04:54,337 --> 00:04:56,473
AND HIS DOCTOR OF MEDICINE AND 

127
00:04:56,473 --> 00:04:58,041
MASTER OF SCIENCE IN CLINICAL 

128
00:04:58,041 --> 00:04:58,875
INVESTIGATION DEGREES FROM 

129
00:04:58,875 --> 00:04:59,509
WASHINGTON UNIVERSITY SCHOOL OF 

130
00:04:59,509 --> 00:05:00,710
MEDICINE IN ST. LOUIS, MISSOURI.

131
00:05:00,710 --> 00:05:01,311
HIS POST DOCTORAL WORK WAS AT 

132
00:05:01,311 --> 00:05:01,912
WASHINGTON UNIVERSITY IN ST. 

133
00:05:01,912 --> 00:05:02,545
LOUIS, MOORFIELDS EYE HOSPITAL, 

134
00:05:02,545 --> 00:05:03,880
AND THE UNIVERSITY OF BRITISH 

135
00:05:03,880 --> 00:05:05,081
COLUMBIA.

136
00:05:05,081 --> 00:05:05,982
DR. LEE IS AUTHOR OF MORE THAN 

137
00:05:05,982 --> 00:05:07,050
175 PUBLISHED, PEER-REVIEWED 

138
00:05:07,050 --> 00:05:08,118
MANUSCRIPTS, AN ASSOCIATE EDITOR

139
00:05:08,118 --> 00:05:08,718
FOR BOTH TRANSLATIONAL VISION 

140
00:05:08,718 --> 00:05:10,887
SCIENCE AND TECHNOLOGY AND 

141
00:05:10,887 --> 00:05:11,488
OPHTHALMOLOGY SCIENCE, AND AN 

142
00:05:11,488 --> 00:05:12,322
EDITORIAL BOARD MEMBER FOR THE 

143
00:05:12,322 --> 00:05:12,856
AMERICAN JOURNAL OF 

144
00:05:12,856 --> 00:05:16,893
OPHTHALMOLOGY AND NATURE 

145
00:05:16,893 --> 00:05:17,928
SCIENTIFIC REPORTS.

146
00:05:17,928 --> 00:05:20,397
HE IS ALSO THE CHAIR OF THE

147
00:05:20,397 --> 00:05:21,064
AMERICAN ACADEMY OF 

148
00:05:21,064 --> 00:05:21,665
OPHTHALMOLOGY INFORMATION 

149
00:05:21,665 --> 00:05:23,266
TECHNOLOGY STEERING COMMITTEE 

150
00:05:23,266 --> 00:05:25,068
AND HAS RECEIVED MULTIPLE AWARDS

151
00:05:25,068 --> 00:05:28,872
IN HIS FIELD.

152
00:05:28,872 --> 00:05:29,673
TO DATE.

153
00:05:29,673 --> 00:05:30,740
SUPPORTERS OF HIS RESEARCH 

154
00:05:30,740 --> 00:05:31,775
INCLUDE THE NIH COMMON FUND AND 

155
00:05:31,775 --> 00:05:35,612
THE NATIONAL EYE INSTITUTE, AND 

156
00:05:35,612 --> 00:05:37,080
SUPPORT FROM FOUNDATIONS AND 

157
00:05:37,080 --> 00:05:38,081
INDUSTRY.

158
00:05:38,081 --> 00:05:40,450
DR. LEE'S RESEARCH FOCUSES ON 

159
00:05:40,450 --> 00:05:42,385
THE TRANSLATION OF NOVEL 

160
00:05:42,385 --> 00:05:45,588
COMPUTATIONAL TECHNIQUES AND 

161
00:05:45,588 --> 00:05:47,290
MACHINE LEARNING TO UNCOVER NEW 

162
00:05:47,290 --> 00:05:49,826
DISEASE ASSOCIATIONS AND 

163
00:05:49,826 --> 00:05:51,995
MECHANISMS FROM CLINICAL DATA.

164
00:05:51,995 --> 00:05:53,663
IMPORTANTLY, HE IS INTERESTED IN

165
00:05:53,663 --> 00:05:55,632
THE INTERSECTION OF LARGE 

166
00:05:55,632 --> 00:05:57,734
CLINICAL MEDICAL DATASETS AND 

167
00:05:57,734 --> 00:06:00,670
USING NONTRADITIONAL 

168
00:06:00,670 --> 00:06:02,172
COMPUTATIONAL TECHNIQUES TO 

169
00:06:02,172 --> 00:06:03,373
ANALYZE AND VISUALIZE THOSE 

170
00:06:03,373 --> 00:06:03,606
RESULTS.

171
00:06:03,606 --> 00:06:05,709
IN THIS REGARD HE HAS CREATED 

172
00:06:05,709 --> 00:06:07,677
PROGRAMS TO PROCESS NEXT 

173
00:06:07,677 --> 00:06:10,513
GENERATION SEQUENCING DATA IN 

174
00:06:10,513 --> 00:06:12,816
SUPERCOMPUTING ENVIRONMENTS AND 

175
00:06:12,816 --> 00:06:14,017
ANALYZE NUMEROUS BIG DATA 

176
00:06:14,017 --> 00:06:15,919
SOURCES FROM CMS, THE U.S. 

177
00:06:15,919 --> 00:06:17,487
CENSUS AND THE NATIONAL LIBRARY 

178
00:06:17,487 --> 00:06:19,756
OF MEDICINE, MEDLINE ARCHIVES 

179
00:06:19,756 --> 00:06:21,691
JUST TO NAME A FEW.

180
00:06:21,691 --> 00:06:25,095
DR. LEE IS ALSO CO-PRINCIPLE 

181
00:06:25,095 --> 00:06:31,101
INVESTIGATOR FOR THE AI-READI 

182
00:06:31,101 --> 00:06:33,436
EQUITABLE INSIGHTS FONDLY KNOWN 

183
00:06:33,436 --> 00:06:36,973
AS AI-READI SUPPORTED BY THE NIH

184
00:06:36,973 --> 00:06:44,047
COMMON FUND THROUGH THE BRIDGE 

185
00:06:44,047 --> 00:06:46,816
TO AI PROGRAM.

186
00:06:46,816 --> 00:06:48,852
COLLECTING DATA FROM A DIVERSE 

187
00:06:48,852 --> 00:06:50,820
POPULATION SIMULTANEOUSLY 

188
00:06:50,820 --> 00:06:54,090
CREATED A ROAD MAP FOR EQUITABLE

189
00:06:54,090 --> 00:06:55,825
AND EQUITABLE RESEARCH FOCUSING 

190
00:06:55,825 --> 00:06:58,862
ON MORE THAN 4,000 RESEARCH 

191
00:06:58,862 --> 00:07:00,764
PARTICIPANTS TO BETTER 

192
00:07:00,764 --> 00:07:02,999
INVESTIGATE HEALTH OUTOL COMES 

193
00:07:02,999 --> 00:07:06,469
FROM TYPE 2 DIABETES.

194
00:07:06,469 --> 00:07:08,204
IN PREVIOUSLY UNDER SERVED 

195
00:07:08,204 --> 00:07:10,240
POPULATIONS, MANY OF WHOM HAVE 

196
00:07:10,240 --> 00:07:11,608
BEEN IMPACTED BY THIS DISEASE.

197
00:07:11,608 --> 00:07:12,809
IT IS MY PLEASURE TO WELCOME 

198
00:07:12,809 --> 00:07:13,576
YOU, AARON.

199
00:07:13,576 --> 00:07:15,345
I WILL TURN THE MIC OVER TO YOU.

200
00:07:15,345 --> 00:07:17,080
>> THANK YOU SO MUCH,
DAVID.

201
00:07:17,080 --> 00:07:19,582
I ALMOST FEEL LIKE I DON'T NEETD

202
00:07:19,582 --> 00:07:21,951
TO GIVE MY TALK AFTER THAT LONG 

203
00:07:21,951 --> 00:07:22,152
INTRO.

204
00:07:22,152 --> 00:07:28,425
LET ME START SHARING MY SCREEN.

205
00:07:28,425 --> 00:07:32,328
SO IT IS A GREAT HONOR AND A 

206
00:07:32,328 --> 00:07:33,630
PLEASURE TO BE GIVEN THIS 

207
00:07:33,630 --> 00:07:35,632
OPPORTUNITY TO PRESENT TO YOU 

208
00:07:35,632 --> 00:07:35,799
ALL.

209
00:07:35,799 --> 00:07:40,703
ESPECIALLY GIMVEN THAT IT IS TH 

210
00:07:40,703 --> 00:07:42,238
25th ANNIVERSARY OF THE FOUNDING

211
00:07:42,238 --> 00:07:44,140
OF THE NCCIH.

212
00:07:44,140 --> 00:07:44,841
I REALLY APPRECIATE THIS 

213
00:07:44,841 --> 00:07:45,508
OPPORTUNITY.

214
00:07:45,508 --> 00:07:47,310
COMING FROM A DIFFERENT FIELD, 

215
00:07:47,310 --> 00:07:49,612
PLEASE FORGIVE ME IF I MAKE A 

216
00:07:49,612 --> 00:07:51,481
MISTAKE OR SAY SOMETHING 

217
00:07:51,481 --> 00:07:51,781
INCORRECTLY.

218
00:07:51,781 --> 00:07:53,550
I'M MUCH MORE FAMILIAR WITH THE 

219
00:07:53,550 --> 00:07:57,087
FIELD OF OPHTHALMOLOGY THAN I AM

220
00:07:57,087 --> 00:07:57,987
IN COMPLEMENTARY MEDICINE.

221
00:07:57,987 --> 00:08:00,523
SO I JUST WANT TO START BY 

222
00:08:00,523 --> 00:08:02,325
SHOWING MY FINANCIAL 

223
00:08:02,325 --> 00:08:02,692
DISCLOSURES.

224
00:08:02,692 --> 00:08:05,328
SOME OF THESE COMPANIES ARE 

225
00:08:05,328 --> 00:08:07,597
SUPPORTING THE WORK IN AI-READI 

226
00:08:07,597 --> 00:08:09,866
SO THERE IS SOME POTENTIAL 

227
00:08:09,866 --> 00:08:11,034
CONFLICTS OF INTEREST.

228
00:08:11,034 --> 00:08:14,003
SO ROUGHLY IN MY TALK IS DIVIDED

229
00:08:14,003 --> 00:08:15,638
INTO THESE SORT OF FOUR 

230
00:08:15,638 --> 00:08:15,905
SECTIONS.

231
00:08:15,905 --> 00:08:17,307
I'M GOING TO START BY GIVING YOU

232
00:08:17,307 --> 00:08:20,743
A BIT OF AN INTRODUCTION TO AI 

233
00:08:20,743 --> 00:08:22,979
AND SPECIFICALLY DEEP LEARNING.

234
00:08:22,979 --> 00:08:27,283
AND THEN WALK YOU THROUGH MY 

235
00:08:27,283 --> 00:08:28,718
JOURNEY USING DEEP LEARNING 

236
00:08:28,718 --> 00:08:30,920
TECHNIQUES IN THE FIELD OF 

237
00:08:30,920 --> 00:08:31,254
OPHTHALMOLOGY.

238
00:08:31,254 --> 00:08:34,524
THEN I'M GOING TO TALK A BIT 

239
00:08:34,524 --> 00:08:37,494
ABOUT MANIFOLD LEARNING AND HOW 

240
00:08:37,494 --> 00:08:39,429
THAT LED US TO CONCEIVE OF THE 

241
00:08:39,429 --> 00:08:41,564
AI-READI PROJECT THAT IS PART OF

242
00:08:41,564 --> 00:08:43,166
THE BRIDGE TO AI PROGRAM.

243
00:08:43,166 --> 00:08:45,034
I JUST WANT TO START BY 

244
00:08:45,034 --> 00:08:46,870
GROUNDING EVERYBODY IN WHERE WE 

245
00:08:46,870 --> 00:08:48,605
ARE AS A SOCIETY.

246
00:08:48,605 --> 00:08:50,640
YOU KNOW, WE'RE REALLY LIVING IN

247
00:08:50,640 --> 00:08:52,876
A BRAND NEW ERA WHERE DATA IS 

248
00:08:52,876 --> 00:08:56,713
BEING CREATED ALL AROUND US AND 

249
00:08:56,713 --> 00:08:58,181
OUR ALGORITHMS THAT ARE WATCHING

250
00:08:58,181 --> 00:09:02,585
AND HARNESSING THAT DATA AND 

251
00:09:02,585 --> 00:09:03,119
INFLUENCING PRETTY MUCH 

252
00:09:03,119 --> 00:09:04,821
EVERYTHING WE SEE AND DO.

253
00:09:04,821 --> 00:09:07,223
EVERY TIME YOU OPEN YOUR PHONE, 

254
00:09:07,223 --> 00:09:08,992
TYPE SOMETHING INTO GOOGLE, 

255
00:09:08,992 --> 00:09:11,361
COMPOSE AN E-MAIL, ALGORITHMS 

256
00:09:11,361 --> 00:09:12,061
ARE WATCHING AND LEARNING FROM 

257
00:09:12,061 --> 00:09:13,730
THAT DATA.

258
00:09:13,730 --> 00:09:16,466
ONE OF THE THINGS THAT I REALLY 

259
00:09:16,466 --> 00:09:20,803
LIKE TO DO IS DISAMBIGUATE ABOUT

260
00:09:20,803 --> 00:09:22,405
THESE TERMS.

261
00:09:22,405 --> 00:09:25,642
THE NEWS MEDIA AND SOMETIMES THE

262
00:09:25,642 --> 00:09:27,644
MEDICAL INDUSTRY USING THEM 

263
00:09:27,644 --> 00:09:29,445
INTERCHANGEABLE, BUT THEY ARE 

264
00:09:29,445 --> 00:09:30,947
ACTUALLY SEPARATE FIELDS AND 

265
00:09:30,947 --> 00:09:31,347
DEFINITIONS.

266
00:09:31,347 --> 00:09:32,315
THE FIELD OF ARTIFICIAL 

267
00:09:32,315 --> 00:09:34,117
INTELLIGENCE IS ACTUALLY QUITE 

268
00:09:34,117 --> 00:09:34,450
OLD.

269
00:09:34,450 --> 00:09:36,586
IT HAS EXISTED SINCE THE DAWN OF

270
00:09:36,586 --> 00:09:37,854
MODERN COMPUTING.

271
00:09:37,854 --> 00:09:39,255
THERE IS A SUB FIELD WITHIN 

272
00:09:39,255 --> 00:09:41,057
THERE KNOWN AS MACHINE LEARNING.

273
00:09:41,057 --> 00:09:42,926
AND THEN WITHIN THE FIELD OF 

274
00:09:42,926 --> 00:09:44,093
MACHINE LEARNING THERE IS A SUB 

275
00:09:44,093 --> 00:09:45,395
FIELD THAT IS RELATIVELY NEW 

276
00:09:45,395 --> 00:09:47,530
CALLED DEEP LEARNING.

277
00:09:47,530 --> 00:09:49,098
AND TO UNDERSTAND WHAT DEEP 

278
00:09:49,098 --> 00:09:51,034
LEARNING IS, YOU HAVE TO 

279
00:09:51,034 --> 00:09:52,869
UNDERSTAND WHAT AN ARTIFICIAL 

280
00:09:52,869 --> 00:09:53,836
NEURAL NETWORK IS.

281
00:09:53,836 --> 00:09:56,439
THESE WERE VERY POPULAR IN THE 

282
00:09:56,439 --> 00:09:58,274
1980s AND 1990s WHERE THEY 

283
00:09:58,274 --> 00:10:00,843
THOUGHT THIS WAS GOING TO BE -- 

284
00:10:00,843 --> 00:10:03,546
THIS WAS GOING TO TRANSFORM 

285
00:10:03,546 --> 00:10:04,747
MEDICINE IN EVERYTHING THAT WE 

286
00:10:04,747 --> 00:10:04,914
DO.

287
00:10:04,914 --> 00:10:08,284
SO ALL THE TALK WE ARE SEEING 

288
00:10:08,284 --> 00:10:12,021
TODAY ABOUT AI ACTUALLY, THERE 

289
00:10:12,021 --> 00:10:14,691
WAS ANOTHER AI SUMMER THAT 

290
00:10:14,691 --> 00:10:17,860
OCCURRED IN THE EARLY 90s, LATE 

291
00:10:17,860 --> 00:10:20,196
80s WHERE THEY THOUGHT THESE 

292
00:10:20,196 --> 00:10:21,831
TECHNIQUES WERE GOING TO 

293
00:10:21,831 --> 00:10:22,098
TRANSFORM.

294
00:10:22,098 --> 00:10:24,400
WHEN THE HYPE AND EXPECTATIONS 

295
00:10:24,400 --> 00:10:26,469
DIDN'T MEET REALITY WE WENT INTO

296
00:10:26,469 --> 00:10:27,770
ANNAA WINTER THAT WE ARE COMING 

297
00:10:27,770 --> 00:10:30,073
OUT OF IN RECENT YEARS.

298
00:10:30,073 --> 00:10:31,674
IN MY OPINION, ALL OF THIS IS 

299
00:10:31,674 --> 00:10:33,543
BEING DRIVEN BY THE ADVANCES OF 

300
00:10:33,543 --> 00:10:34,644
DEEP LEARNING.

301
00:10:34,644 --> 00:10:36,379
AND THERE'S FOUR SORT OF THINGS 

302
00:10:36,379 --> 00:10:37,447
THAT HAPPEN AROUND THE SAME TIME

303
00:10:37,447 --> 00:10:40,350
THAT LED TO THE BIRTH OF WHAT WE

304
00:10:40,350 --> 00:10:42,051
CALL DEEP LEARNING TODAY.

305
00:10:42,051 --> 00:10:44,954
THE FIRST WAS THE REALIZATION 

306
00:10:44,954 --> 00:10:47,190
THAT GRAPHICS CARDS THAT EXISTED

307
00:10:47,190 --> 00:10:50,660
IN COMPUTERS COULD BE USED FOR 

308
00:10:50,660 --> 00:10:51,861
ACCELERATING LINEAR ALGEBRA 

309
00:10:51,861 --> 00:10:52,161
OPERATIONS.

310
00:10:52,161 --> 00:10:56,232
THE SECOND IS THERE WERE THESE 

311
00:10:56,232 --> 00:10:57,700
KERNELS THAT WERE SPATIALLY 

312
00:10:57,700 --> 00:11:00,069
AWARE THAT COULD TAKE ADVANTAGE 

313
00:11:00,069 --> 00:11:02,071
OF DATA THAT WERE SPATIALLY 

314
00:11:02,071 --> 00:11:03,239
CORRELATED WITH ONE ANOTHER.

315
00:11:03,239 --> 00:11:05,375
WHAT I MEAN BY THAT IS THINGS 

316
00:11:05,375 --> 00:11:09,078
LIKE IMAGES OR THINGS LIKE 

317
00:11:09,078 --> 00:11:11,381
SEQUENCES OF WORDS.

318
00:11:11,381 --> 00:11:13,983
THERE'S PATTERNS THAT OCCUR WITH

319
00:11:13,983 --> 00:11:16,052
NEIGHBORING TOKENS OR PIXELS 

320
00:11:16,052 --> 00:11:17,920
THAT LED TO INFORMATION THAT THE

321
00:11:17,920 --> 00:11:19,489
MODELS COULD EXPLOIT.

322
00:11:19,489 --> 00:11:21,057
THE THIRD WAS THE USE OF 

323
00:11:21,057 --> 00:11:24,193
SOMETHING CALLED NONLINEAR 

324
00:11:24,193 --> 00:11:27,163
ACTIVATION FUNCTIONS AND THEN 

325
00:11:27,163 --> 00:11:28,064
FINALLY, YOU KNOW, OUR COMPUTER 

326
00:11:28,064 --> 00:11:29,632
HARDWARE AND TECHNIQUES HAD TO 

327
00:11:29,632 --> 00:11:35,071
EVOLVE TO A POINT WHERE WE COULD

328
00:11:35,071 --> 00:11:38,274
COLLECT AND COLATE LARGE AMOUNTS

329
00:11:38,274 --> 00:11:39,709
OF DATA.

330
00:11:39,709 --> 00:11:40,877
PUTTING THESE THINGS TOGETHER WE

331
00:11:40,877 --> 00:11:42,912
LIVE IN A VERY EXCITING TIME AS 

332
00:11:42,912 --> 00:11:45,081
WE START TO PUSH THESE ALGOR

333
00:11:45,081 --> 00:11:47,684
ALGORITHMS WITH MORE AND MORE 

334
00:11:47,684 --> 00:11:49,886
DATA AND MAKE THE MODELS LARGER 

335
00:11:49,886 --> 00:11:51,888
AND LARGER THEIRING PERFORMANCE 

336
00:11:51,888 --> 00:11:55,325
SEEMS TO INCREASE AND INCREASE.

337
00:11:55,325 --> 00:11:57,960
THIS LED TO WHAT PEOPLE ARE 

338
00:11:57,960 --> 00:11:59,195
CALLED THE FOURTH INDUSTRIAL 

339
00:11:59,195 --> 00:12:00,496
REVOLUTION WE ARE LIVING THROUGH

340
00:12:00,496 --> 00:12:01,064
TODAY.

341
00:12:01,064 --> 00:12:03,199
THE CONVERGENCE OF BOTH THE 

342
00:12:03,199 --> 00:12:04,867
DATASETS LARGE ENOUGH TO TRAIN 

343
00:12:04,867 --> 00:12:07,603
THE MODELS AS WELL AS THE 

344
00:12:07,603 --> 00:12:08,871
TECHNIQUES THAT CAN MAKE THEM 

345
00:12:08,871 --> 00:12:10,073
HAPPEN IN A REASONABLE TIME.

346
00:12:10,073 --> 00:12:12,108
THAT IS LEADING TO ALL THE 

347
00:12:12,108 --> 00:12:14,243
THINGS WE ARE READING ABOUT 

348
00:12:14,243 --> 00:12:18,247
TODAY, SELF-DRIVING CARS, 

349
00:12:18,247 --> 00:12:19,982
CHATGPT, ALL OF THESE ARE BORN 

350
00:12:19,982 --> 00:12:21,651
OUT OF THIS CONVERGENCE NOW.

351
00:12:21,651 --> 00:12:23,920
I WANT TO REWIND THE CLOCK A 

352
00:12:23,920 --> 00:12:25,088
LITTLE BIT AND TELL YOU ABOUT 

353
00:12:25,088 --> 00:12:26,322
HOW WE GOT STAFF REPORTED IN 

354
00:12:26,322 --> 00:12:26,856
THIS.

355
00:12:26,856 --> 00:12:28,858
TO UNDERSTAND THIS YOU SORT OF 

356
00:12:28,858 --> 00:12:30,860
HAVE TO UNDERSTAND A LITTLE BIT 

357
00:12:30,860 --> 00:12:32,061
ABOUT OUR FIELD.

358
00:12:32,061 --> 00:12:34,731
THERE'S AN IMAGING MODALITY THAT

359
00:12:34,731 --> 00:12:43,306
WE USE CALLED OPTICAL COHERENCE 

360
00:12:43,306 --> 00:12:43,706
TOMOLOGY.

361
00:12:43,706 --> 00:12:45,141
THIS IS A GRAY SCALE WHERE IT IS

362
00:12:45,141 --> 00:12:46,709
A CROSS SECTION OF THE RETINA 

363
00:12:46,709 --> 00:12:48,878
AND IT SHOWS YOU THE 

364
00:12:48,878 --> 00:12:50,646
ARCHITECTURE OF THE RETINA IN A 

365
00:12:50,646 --> 00:12:52,782
WAY YOU CAN'T REALLY SEE WITH 

366
00:12:52,782 --> 00:12:54,751
THE NAKED EYE.

367
00:12:54,751 --> 00:12:57,720
WE AS CLINICIANS, WE LOOK INSIDE

368
00:12:57,720 --> 00:13:00,123
PEOPLE'S EYES WE ARE ABLE TO SEE

369
00:13:00,123 --> 00:13:02,725
PICTURES THAT LOOK LIKE THIS.

370
00:13:02,725 --> 00:13:04,927
MOST OF THE PEOPLE OUTSIDE OF 

371
00:13:04,927 --> 00:13:05,962
THE VISION SCIENCE ASSUME WHEN 

372
00:13:05,962 --> 00:13:12,869
WE ARE TALKING ABOUT RETIANL 

373
00:13:12,869 --> 00:13:15,071
IMAGES AND PHOTOGRAPHY, WE ARE 

374
00:13:15,071 --> 00:13:16,606
TALKING ABOUT THE IMAGES ON THE 

375
00:13:16,606 --> 00:13:16,939
RIGHT.

376
00:13:16,939 --> 00:13:22,912
THE DATA THAT IS GENERATED AND 

377
00:13:22,912 --> 00:13:24,380
AVAILABLE FOR RESEARCH IS THESE 

378
00:13:24,380 --> 00:13:26,015
IMAGES ON THE LEFT.

379
00:13:26,015 --> 00:13:28,518
THIS IMAGING MODALITY ALLOWS US 

380
00:13:28,518 --> 00:13:31,687
TO SEE ALMOST A CELLULAR LEVEL 

381
00:13:31,687 --> 00:13:33,156
INFORMATION NONINVASIONIVELY.

382
00:13:33,156 --> 00:13:38,528
IT ALLOWS US TO SEE, THIS IS A 

383
00:13:38,528 --> 00:13:39,996
HISTOLOGY SLIDE OF THE HUMAN 

384
00:13:39,996 --> 00:13:40,229
RETINA.

385
00:13:40,229 --> 00:13:47,703
YOU CAN SEE THE STRIATIONS THAN 

386
00:13:47,703 --> 00:13:49,539
YOU CAN ON A PATHOLOGY SLIDE.

387
00:13:49,539 --> 00:13:52,308
WE DIDN'T HAVE TO TAKE A PATH 

388
00:13:52,308 --> 00:13:53,709
SLIDE OUT OF SOMEBODY WHO IS 

389
00:13:53,709 --> 00:13:55,344
DECEASED TO TAKE THIS IMAGE.

390
00:13:55,344 --> 00:13:58,948
WE CAN TAKE IT IN LIVE PEOPLE.

391
00:13:58,948 --> 00:14:01,517
THE OCT SCAN CAN BE THOUGHT OF 

392
00:14:01,517 --> 00:14:03,886
AS A CT SCAN OF THE MACULA.

393
00:14:03,886 --> 00:14:07,590
IT PROVIDES A THREE-DIMENSIONAL 

394
00:14:07,590 --> 00:14:08,691
INFORMATION ABOUT EVERY POSITION

395
00:14:08,691 --> 00:14:11,360
IN THE MACULA.

396
00:14:11,360 --> 00:14:14,130
AND WE, BECAUSE IT ALLOWS US TO 

397
00:14:14,130 --> 00:14:16,833
SEE THINGS WE CAN'T NORMALLY 

398
00:14:16,833 --> 00:14:19,335
SEE, WE HAVE BEEN DOING A LOT OF

399
00:14:19,335 --> 00:14:19,535
THEM.

400
00:14:19,535 --> 00:14:21,704
THIS IS DATA FROM A NUMBER OF 

401
00:14:21,704 --> 00:14:22,472
YEARS AGO NOW.

402
00:14:22,472 --> 00:14:24,307
JUST IN THE UNITED STATES ALONE,

403
00:14:24,307 --> 00:14:26,108
WE ARE BREAKING 5 MILLION SCANS 

404
00:14:26,108 --> 00:14:27,243
A YEAR.

405
00:14:27,243 --> 00:14:29,846
AND IT IS BY FAR ONE OF THE MOST

406
00:14:29,846 --> 00:14:32,915
COMMON IMAGING MODALITIES 

407
00:14:32,915 --> 00:14:36,586
CAPTURED IN MEDICINE OVERALL.

408
00:14:36,586 --> 00:14:40,022
SO SO OUR FIRST FORRAY INTO DEEP

409
00:14:40,022 --> 00:14:41,858
LEARNING CAME FROM THIS DATASET 

410
00:14:41,858 --> 00:14:43,860
I CREATED WHEN I STARTED AT THE 

411
00:14:43,860 --> 00:14:46,963
UNIVERSITY OF WASHINGTON.

412
00:14:46,963 --> 00:14:49,065
IT HAD ABOUT 5.5 MILLION OF 

413
00:14:49,065 --> 00:14:50,066
THESE IMAGES.

414
00:14:50,066 --> 00:14:53,503
BECAUSE WE HAD ACCESS TO THE 

415
00:14:53,503 --> 00:14:53,669
EHR.

416
00:14:53,669 --> 00:14:56,372
WE COULD EXTRACT THESE VARIABLES

417
00:14:56,372 --> 00:14:57,940
AND MATCH THEM UP.

418
00:14:57,940 --> 00:14:59,375
THE VERY FIRST QUESTION WE SORT 

419
00:14:59,375 --> 00:15:01,777
OF ASKED WAS SORT OF THE, HELLO,

420
00:15:01,777 --> 00:15:02,845
WORLD, EQUIVALENT OF DEEP 

421
00:15:02,845 --> 00:15:05,515
LEARNING WHERE IT IS A VERY 

422
00:15:05,515 --> 00:15:07,717
SIMPLE, VERY EASY CLINICAL 

423
00:15:07,717 --> 00:15:08,017
PROBLEM.

424
00:15:08,017 --> 00:15:10,319
AND THAT'S THIS QUESTION OF 

425
00:15:10,319 --> 00:15:11,487
WHETHER, YOU KNOW, A DEEP 

426
00:15:11,487 --> 00:15:12,755
LEARNING MODEL COULD DISTINGUISH

427
00:15:12,755 --> 00:15:18,227
BETWEEN A NORMAL SCAN AND AN AMD

428
00:15:18,227 --> 00:15:18,661
SCAN.

429
00:15:18,661 --> 00:15:21,464
AGE RELATED MACULAR 

430
00:15:21,464 --> 00:15:21,797
DEGENERATION.

431
00:15:21,797 --> 00:15:23,466
WHICH IS ONE OF THE MOST COMMON 

432
00:15:23,466 --> 00:15:25,401
REASONS WE OBTAINED THESE 

433
00:15:25,401 --> 00:15:25,701
IMAGES.

434
00:15:25,701 --> 00:15:28,137
WITH OUR DATASET WE HAD ROUGHLY 

435
00:15:28,137 --> 00:15:31,307
100,000 IMAGES THAT WE TRAINED.

436
00:15:31,307 --> 00:15:33,376
THIS MODEL STATE OF THE ARD 

437
00:15:33,376 --> 00:15:37,179
CALLED VGE16 AND WITHIN A COUPLE

438
00:15:37,179 --> 00:15:39,248
OF DAYS WE WERE ABLE TO SHOW 

439
00:15:39,248 --> 00:15:42,952
THIS KIND OF PERFORMANCE.

440
00:15:42,952 --> 00:15:45,087
WE WERE ABLE TO JUST FROM GIVING

441
00:15:45,087 --> 00:15:47,790
THE MODEL THESE IMAGES OF THE 

442
00:15:47,790 --> 00:15:49,392
OCT SCANS AS WELL AS THESE 

443
00:15:49,392 --> 00:15:51,360
PEOPLE HAD AMD AND THESE PEOPLE 

444
00:15:51,360 --> 00:15:52,595
WERE NORMAL.

445
00:15:52,595 --> 00:15:54,697
THE MODEL COULD LEARN TO 

446
00:15:54,697 --> 00:15:56,232
DISTINGUISH BETWEEN THESE TWO 

447
00:15:56,232 --> 00:15:58,167
WITH ALMOST NEAR PERFECTION.

448
00:15:58,167 --> 00:16:00,670
AND THIS WAS SORT OF ASTONISHING

449
00:16:00,670 --> 00:16:02,939
THAT THE MODEL WAS ABLE TO DO 

450
00:16:02,939 --> 00:16:05,074
THIS WITHOUT ANY SORT OF HUMAN 

451
00:16:05,074 --> 00:16:10,146
GUIDANCE OF HOW TO ANALYZE THE 

452
00:16:10,146 --> 00:16:10,379
IMAGES.

453
00:16:10,379 --> 00:16:11,914
ONE OF THE THINGS WE DID TO 

454
00:16:11,914 --> 00:16:13,883
PROVE THAT THIS WAS NOT SOME 

455
00:16:13,883 --> 00:16:17,386
FLUKE IS WE PROVIDED SOME POST 

456
00:16:17,386 --> 00:16:18,854
DOC VISUALIZATION THAT LOOKED AT

457
00:16:18,854 --> 00:16:21,290
WHERE THE MODEL WAS MOST 

458
00:16:21,290 --> 00:16:23,292
DEPENDENT TO MAKE THIS 

459
00:16:23,292 --> 00:16:24,393
DETERMINATION AND IT MATCHED 

460
00:16:24,393 --> 00:16:27,096
WITH THE AREAS OF PATHOLOGY THAT

461
00:16:27,096 --> 00:16:30,466
WE ARE USED TO SEEING AS 

462
00:16:30,466 --> 00:16:30,766
CLINICIANS.

463
00:16:30,766 --> 00:16:32,902
THIS GAVE US SOME CERTAINTY THAT

464
00:16:32,902 --> 00:16:36,973
THE MODEL WAS DOING SOMETHING 

465
00:16:36,973 --> 00:16:37,273
REASONABLE.

466
00:16:37,273 --> 00:16:39,675
WE WENT ON AS OUR LAB TO PUBLISH

467
00:16:39,675 --> 00:16:42,878
MANY, MANY PAPERS IN OUR FIELD.

468
00:16:42,878 --> 00:16:45,081
SHOWING THAT DEEP LEARNING 

469
00:16:45,081 --> 00:16:46,949
MODELS COULD DO THINGS THAT 

470
00:16:46,949 --> 00:16:48,451
HUMAN CLINICIANS COULD DO AND 

471
00:16:48,451 --> 00:16:51,854
EVEN SHOW US NEW PATTERNS OF 

472
00:16:51,854 --> 00:16:52,254
DISEASE.

473
00:16:52,254 --> 00:16:54,357
AND SO THIS IS, YOU KNOW, REALLY

474
00:16:54,357 --> 00:16:56,092
THE TAKE HOME MESSAGE OF THIS 

475
00:16:56,092 --> 00:16:56,425
SECTION.

476
00:16:56,425 --> 00:16:58,728
THAT WE ARE LIVING WITH A VERY 

477
00:16:58,728 --> 00:17:01,664
POWERFUL NEW TOOL SET THAT USES 

478
00:17:01,664 --> 00:17:03,132
DEEP LEARNING AND IT HAS THE 

479
00:17:03,132 --> 00:17:04,800
ABILITY TO LEARN DIRECTLY FROM 

480
00:17:04,800 --> 00:17:08,671
THE DATA AND DISCOVER NEW 

481
00:17:08,671 --> 00:17:09,305
PATTERNS.

482
00:17:09,305 --> 00:17:13,209
I WANT TO TALK A BIT ABOUT 

483
00:17:13,209 --> 00:17:18,848
MANIFOLD LEARNING.

484
00:17:18,848 --> 00:17:19,982
WHICH IS A SIMPLE IDEA.

485
00:17:19,982 --> 00:17:22,051
IT COMES FROM THIS IDEA OF 

486
00:17:22,051 --> 00:17:23,386
DIMENSION REDUCTION.

487
00:17:23,386 --> 00:17:25,888
ONE WAY TO THINK ABOUT IMAGE OR 

488
00:17:25,888 --> 00:17:29,225
A DATASET, SOMETHING LIKE WHOLE 

489
00:17:29,225 --> 00:17:31,193
GENOME SEQUENCING OR A WHOLE 

490
00:17:31,193 --> 00:17:33,295
BUNCH OF SURVEY RESULTS IS THAT 

491
00:17:33,295 --> 00:17:34,897
THOSE DATA ELEMENTS, WHEN THEY 

492
00:17:34,897 --> 00:17:37,033
ARE BROKEN DOWN PER PARTICIPANT,

493
00:17:37,033 --> 00:17:43,506
LIVE IN A VERY HIGH DIMENSIONAL 

494
00:17:43,506 --> 00:17:43,706
SPACE.

495
00:17:43,706 --> 00:17:45,875
WHICH IS NOT VERY USEFUL FOR 

496
00:17:45,875 --> 00:17:46,242
ANALYSIS.

497
00:17:46,242 --> 00:17:48,344
WE REALLY WANT TO REDUCE IT DOWN

498
00:17:48,344 --> 00:17:50,613
TO SOMETHING AS SIMPLE AS TWO 

499
00:17:50,613 --> 00:17:51,781
DIMENSIONS WHERE WE CAN 

500
00:17:51,781 --> 00:17:53,215
UNDERSTAND THE PATTERNS AND 

501
00:17:53,215 --> 00:17:56,252
RELATIONSHIPS THAT ARE IN THE 

502
00:17:56,252 --> 00:17:56,786
DATASET.

503
00:17:56,786 --> 00:17:59,488
SO HERE THIS LATENT SPACE, THIS 

504
00:17:59,488 --> 00:18:01,490
LOWER DIMENSIONAL LATENT SPACE.

505
00:18:01,490 --> 00:18:03,459
IS A TWO-DIMENSIONAL GRAPH WHERE

506
00:18:03,459 --> 00:18:05,327
IT OBEYS TWO PRINCIPLES.

507
00:18:05,327 --> 00:18:07,430
TWO POINTS THAT ARE CLOSE 

508
00:18:07,430 --> 00:18:09,065
TOGETHER ON THIS SPACE ARE VERY 

509
00:18:09,065 --> 00:18:10,766
SIMILAR TO ONE ANOTHER.

510
00:18:10,766 --> 00:18:13,102
AND TWO POINTS THAT ARE FAR 

511
00:18:13,102 --> 00:18:14,270
APART ARE VERY DIFFERENT.

512
00:18:14,270 --> 00:18:16,439
AND THAT IS THE ONLY REAL 

513
00:18:16,439 --> 00:18:18,340
ORGANIZATIONAL PRINCIPLE.

514
00:18:18,340 --> 00:18:20,376
AND IF WE CAN SOMEHOW DISTILL 

515
00:18:20,376 --> 00:18:22,511
THE DATA DOWN IN A WAY THAT 

516
00:18:22,511 --> 00:18:24,613
OBEYS THESE TWO PRINCIPLES, WE 

517
00:18:24,613 --> 00:18:26,082
CAN START TO ORGANIZE AND MAP 

518
00:18:26,082 --> 00:18:28,784
THAT DATA IN A WAY THAT WE CAN 

519
00:18:28,784 --> 00:18:30,586
UNDERSTAND IT.

520
00:18:30,586 --> 00:18:32,855
THERE'S A CLASS OF DEEP LEARNING

521
00:18:32,855 --> 00:18:34,957
METHODS KNOWN AS SELF-SUPERVISED

522
00:18:34,957 --> 00:18:36,759
LEARNING AND THIS IS AN 

523
00:18:36,759 --> 00:18:38,327
ANIMATION TAKEN FROM FACEBOOK 

524
00:18:38,327 --> 00:18:41,097
WHERE THEY DEVELOPED A MODEL 

525
00:18:41,097 --> 00:18:44,133
CALLED DINO, USING THIS 

526
00:18:44,133 --> 00:18:45,234
FRAMEWORK WHERE THERE'S NO 

527
00:18:45,234 --> 00:18:46,736
LABELS HERE.

528
00:18:46,736 --> 00:18:47,970
THERE'S NO INFORMATION ABOUT 

529
00:18:47,970 --> 00:18:49,972
WHAT THESE IMAGES ARE, BUT JUST 

530
00:18:49,972 --> 00:18:51,240
FROM THE IDEA THAT A DEEP 

531
00:18:51,240 --> 00:18:52,842
LEARNING MODEL SHOULD BE 

532
00:18:52,842 --> 00:18:54,043
CONSISTENT, IT IS ENOUGH TO 

533
00:18:54,043 --> 00:18:58,714
TRAIN THESE MODELS IN A WAY THAT

534
00:18:58,714 --> 00:18:59,749
CAN EXTRACT INFORMATION.

535
00:18:59,749 --> 00:19:01,951
IT IS TO THE POINT WHERE THESE 

536
00:19:01,951 --> 00:19:03,786
MODELS ARE ACTUALLY ABLE TO TAKE

537
00:19:03,786 --> 00:19:06,655
A SET OF GENERIC IMAGES WITH NO 

538
00:19:06,655 --> 00:19:08,924
LABELS ATTACHED TO THEM AND 

539
00:19:08,924 --> 00:19:10,693
ORGANIZE THEM INTO THESE 

540
00:19:10,693 --> 00:19:12,361
CLUSTERS, INTO THIS LOWER 

541
00:19:12,361 --> 00:19:13,362
DIMENSIONAL SPACE WHERE YOU CAN 

542
00:19:13,362 --> 00:19:16,966
SEE DIFFERENT PICTURES OF CATS 

543
00:19:16,966 --> 00:19:18,968
CLUSTERED TOGETHER, DIFFERENT 

544
00:19:18,968 --> 00:19:21,036
PICTURES OF DOGS CLUSTERED 

545
00:19:21,036 --> 00:19:22,905
TOGETHER, BUT CATS AND DOGS ARE 

546
00:19:22,905 --> 00:19:24,974
FAR APART ON THIS LOWER 

547
00:19:24,974 --> 00:19:25,741
DIMENSIONAL SPACE.

548
00:19:25,741 --> 00:19:27,877
WHAT IS REALLY NEAT ABOUT THIS 

549
00:19:27,877 --> 00:19:29,612
FRAMEWORK, IT CAN THEN TAKE EVEN

550
00:19:29,612 --> 00:19:31,013
MOVIES AND PROCESS THEM AND 

551
00:19:31,013 --> 00:19:33,916
POINT OUT THE MOST RELEVANT 

552
00:19:33,916 --> 00:19:34,750
FOREGROUND OBJECT THAT IS 

553
00:19:34,750 --> 00:19:36,986
PRESENT INSIDE OF THESE VIDEOS.

554
00:19:36,986 --> 00:19:38,687
AND, AGAIN, WHAT IS AMAZING HERE

555
00:19:38,687 --> 00:19:40,990
IS THAT THE MODELS LEARN TO DO 

556
00:19:40,990 --> 00:19:43,793
THIS ON THEIR OWN.

557
00:19:43,793 --> 00:19:46,061
ONE OF OUR COLLABORATIONS WITH 

558
00:19:46,061 --> 00:19:47,496
MICROSOFT WAS TO USE THIS FAMILY

559
00:19:47,496 --> 00:19:49,498
OF TECHNIQUES AND USE THAT 

560
00:19:49,498 --> 00:19:51,700
DATASET THAT I HAD SHOWN YOU 

561
00:19:51,700 --> 00:19:54,303
BEFORE WHERE INSTEAD OF NORMAL 

562
00:19:54,303 --> 00:19:56,572
AMD, WE TOOK THE ENTIRE DATASET 

563
00:19:56,572 --> 00:19:58,974
OF THE 5.5 MILLION AND WE 

564
00:19:58,974 --> 00:20:00,676
APPLIED THESE TECHNIQUES AND 

565
00:20:00,676 --> 00:20:02,711
PROJECTED THEM DOWN INTO A LOWER

566
00:20:02,711 --> 00:20:04,713
DIMENSIONAL SPACE AND THEY 

567
00:20:04,713 --> 00:20:07,049
FORMED THESE SORT OF BEAUTIFUL 

568
00:20:07,049 --> 00:20:09,084
ONE DIMENSIONAL MANIFOLDS.

569
00:20:09,084 --> 00:20:11,854
YOU CAN SEE WHEN YOU COLOR BY 

570
00:20:11,854 --> 00:20:13,923
AGE, THE MANIFOLDS ARE OBEYING 

571
00:20:13,923 --> 00:20:15,925
THE PRINCIPLES PEOPLE ON THIS 

572
00:20:15,925 --> 00:20:16,992
SIDE ARE MUCH YOUNGER AND THE 

573
00:20:16,992 --> 00:20:19,461
PEOPLE ON THE UPPER RIGHT SIDE 

574
00:20:19,461 --> 00:20:21,664
ARE MUCH OLDER AND MOST OF THE 

575
00:20:21,664 --> 00:20:24,366
DISEASES ARE OCCURRING IN THIS 

576
00:20:24,366 --> 00:20:25,301
UPPER RIGHT QUADRANT.

577
00:20:25,301 --> 00:20:27,469
SO THIS IS AN EXAMPLE OF WHERE 

578
00:20:27,469 --> 00:20:29,939
WE'VE BEEN ABLE TO ACTUALLY TAKE

579
00:20:29,939 --> 00:20:32,608
A SINGLE IMAGE AND EMBED THEM IN

580
00:20:32,608 --> 00:20:35,377
THIS LOWER DIMENSIONAL SPACE IN 

581
00:20:35,377 --> 00:20:36,712
A WAY WE CAN UNDERSTAND THAT 

582
00:20:36,712 --> 00:20:37,079
PATTERN.

583
00:20:37,079 --> 00:20:39,215
I WANT TO FOCUS THIS IDEA DOWN 

584
00:20:39,215 --> 00:20:42,918
INTO A SINGLE  DISEASE.

585
00:20:42,918 --> 00:20:44,420
THIS IS TAKING THE SAME IDEA 

586
00:20:44,420 --> 00:20:48,557
EXCEPT WE ARE USING ONE HUMAN 

587
00:20:48,557 --> 00:20:50,292
DISEASE KNOWNED A MACULAR -- WE 

588
00:20:50,292 --> 00:20:52,862
KNOW THE DYNAMICS OF THIS 

589
00:20:52,862 --> 00:20:54,330
DISEASE, THAT IT ONLY REALLY 

590
00:20:54,330 --> 00:20:56,165
AFFECTED THE EYE AND IT IS 

591
00:20:56,165 --> 00:20:57,066
PROGRESSIVE.

592
00:20:57,066 --> 00:20:58,868
ONCE YOU START HAVING THIS 

593
00:20:58,868 --> 00:21:00,603
DISEASE, IT CONTINUES TO 

594
00:21:00,603 --> 00:21:03,005
PROGRESS IN A UNIDIRECTIONAL 

595
00:21:03,005 --> 00:21:03,239
FASHION.

596
00:21:03,239 --> 00:21:04,673
WHAT WE WERE ABLE TO DO IS TRAIN

597
00:21:04,673 --> 00:21:08,277
THIS MODEL IN A WAY THAT WOULD 

598
00:21:08,277 --> 00:21:10,512
CREATE THIS ONE-DIMENSIONAL 

599
00:21:10,512 --> 00:21:12,548
MANIFOLD WHERE THE UPPER LEFT IS

600
00:21:12,548 --> 00:21:14,617
VERY HEALTHY AND NORMAL AND THE 

601
00:21:14,617 --> 00:21:17,519
LOWER RIGHT IS VERY DISEASED.

602
00:21:17,519 --> 00:21:19,255
THIS ONE-DIMENSIONAL MANIFOLD IS

603
00:21:19,255 --> 00:21:21,056
ABLE TO SORT OF SHOW US A 

604
00:21:21,056 --> 00:21:24,193
PROGRESSION OR A SPECTRUM OR A 

605
00:21:24,193 --> 00:21:25,394
CONTINUUM OF DISEASE AS IT GOES 

606
00:21:25,394 --> 00:21:28,998
FROM VERY HEALTHY TO HAVE 

607
00:21:28,998 --> 00:21:29,899
DISEASED.

608
00:21:29,899 --> 00:21:31,634
NOW, WHAT'S REALLY NEAT IS WE 

609
00:21:31,634 --> 00:21:33,969
DEVELOPED THIS SORT OF 

610
00:21:33,969 --> 00:21:35,004
VISUALIZATION TO KIND OF SHOW, 

611
00:21:35,004 --> 00:21:37,072
YOU KNOW, A FLY THROUGH OF THIS 

612
00:21:37,072 --> 00:21:38,774
ONE DIMENSIONAL MANIFOLD AND AS 

613
00:21:38,774 --> 00:21:41,310
YOU CAN SORT OF SEE IN THE 

614
00:21:41,310 --> 00:21:43,746
GRAPHS THAT ARE SHOWING UP, THAT

615
00:21:43,746 --> 00:21:45,447
THERE'S MORE AND MORE DISEASE 

616
00:21:45,447 --> 00:21:47,750
BEING PRESENT IN THOSE OCT 

617
00:21:47,750 --> 00:21:49,985
IMAGES AS WE GO FROM THE UPPER 

618
00:21:49,985 --> 00:21:52,454
LEFT DOWN TO THE LOWER RIGHT.

619
00:21:52,454 --> 00:21:55,524
MY POINT HERE IS REALLY THAT 

620
00:21:55,524 --> 00:21:56,892
DEEP LEARNING, THESE TECHNIQUES,

621
00:21:56,892 --> 00:21:58,827
THESE SETS OF TECHNIQUES THAT 

622
00:21:58,827 --> 00:22:03,565
ARE BECOMING MORE POPULAR CAN 

623
00:22:03,565 --> 00:22:04,533
REALLY SORT OF ORGANIZE 

624
00:22:04,533 --> 00:22:07,636
INFORMATION COMPLETELY DE NOVO, 

625
00:22:07,636 --> 00:22:09,972
SO WITHOUT ANY GUIDANCE FROM THE

626
00:22:09,972 --> 00:22:10,272
CLINICIANS.

627
00:22:10,272 --> 00:22:13,142
THESE MODELS ARE ABLE TO EXTRACT

628
00:22:13,142 --> 00:22:14,143
OUT RELEVANT INFORMATION AND 

629
00:22:14,143 --> 00:22:17,146
ORGANIZE THEM IN A FASHION WHERE

630
00:22:17,146 --> 00:22:18,580
WE CAN START TO UNDERSTAND THE 

631
00:22:18,580 --> 00:22:24,620
NEW PATTERNS OF DISEASE 

632
00:22:24,620 --> 00:22:24,920
PROGRESSION.

633
00:22:24,920 --> 00:22:26,722
SO THIS LAST PART, EVEN THOUGH 

634
00:22:26,722 --> 00:22:28,958
THIS IS THE LAST PART, THIS IS 

635
00:22:28,958 --> 00:22:30,693
THE LONGEST PART OF MY TALK 

636
00:22:30,693 --> 00:22:32,294
BECAUSE I WANTED THOSE OTHER 

637
00:22:32,294 --> 00:22:34,363
SECTIONS TO BE THE BACKDROP TO 

638
00:22:34,363 --> 00:22:36,699
TALKING ABOUT WHAT WE'RE DOING 

639
00:22:36,699 --> 00:22:40,102
IN THE AI-READI PROJECT.

640
00:22:40,102 --> 00:22:42,237
AND TO UNDERSTAND THIS PROGRAM, 

641
00:22:42,237 --> 00:22:43,739
IT'S A LITTLE BIT -- IT IS 

642
00:22:43,739 --> 00:22:45,007
IMPORTANT TO TAKE A STEP BACK 

643
00:22:45,007 --> 00:22:50,045
AND TALK ABOUT THE COMMON FUND 

644
00:22:50,045 --> 00:22:51,046
BRIDGE2AI AM.

645
00:22:51,046 --> 00:22:53,515
THIS IS A PROGRAM HAPPENING NOT 

646
00:22:53,515 --> 00:22:55,951
AT A SINGLE INSTITUTE BUT AT THE

647
00:22:55,951 --> 00:22:58,387
COMMON FUND LEVEL TO TRY TO FILL

648
00:22:58,387 --> 00:23:00,956
THIS BIG GAP THAT EXISTS IN THE 

649
00:23:00,956 --> 00:23:02,424
FIELD OF MEDICINE.

650
00:23:02,424 --> 00:23:04,493
WE REALLY LACK THE AVAILABILITY 

651
00:23:04,493 --> 00:23:09,631
OF WELL-CURATED LARGE DATASETS 

652
00:23:09,631 --> 00:23:11,900
THAT ARE UNBIASED AND FAIR TO 

653
00:23:11,900 --> 00:23:14,603
TRAIN AND DISCOVER NEW THINGS.

654
00:23:14,603 --> 00:23:20,976
SO WE, AS A FIELD, HAVE ONLY 

655
00:23:20,976 --> 00:23:22,711
BEEN ABLE TO USE ROUTINELY 

656
00:23:22,711 --> 00:23:24,913
COLLECTED DATA WITH LOTS OF 

657
00:23:24,913 --> 00:23:27,416
BIASES AND ONLY CAPTURE CERTAIN 

658
00:23:27,416 --> 00:23:28,517
DIMENSIONS OF HEALTH AND NOT 

659
00:23:28,517 --> 00:23:29,818
REALLY CONSIDERED THE WHOLE 

660
00:23:29,818 --> 00:23:34,289
PERSON PICTURE.

661
00:23:34,289 --> 00:23:37,726
SO THE BRIDGE2AI PROGRAM HAS 

662
00:23:37,726 --> 00:23:39,094
THREE PILLARS IT IS ORGANIZED 

663
00:23:39,094 --> 00:23:39,361
AROUND.

664
00:23:39,361 --> 00:23:41,163
FIRST IS THE DATA, THE SECOND IS

665
00:23:41,163 --> 00:23:45,200
ETHICS AND THE THIRD IS THE 

666
00:23:45,200 --> 00:23:45,768
PEOPLE.

667
00:23:45,768 --> 00:23:48,570
AND THE REMIT OF THE PROGRAM, OF

668
00:23:48,570 --> 00:23:50,606
THE BRIDGE2AI IS REALLY TO FOCUS

669
00:23:50,606 --> 00:23:52,441
ON THIS FIRST SORT OF CIRCLE IN 

670
00:23:52,441 --> 00:23:56,645
THE LIFE CYCLE OF, YOU KNOW, 

671
00:23:56,645 --> 00:23:57,679
TRAINING MACHINE LEARNING MODELS

672
00:23:57,679 --> 00:23:59,782
OR DOING AI WORK IN MEDICINE.

673
00:23:59,782 --> 00:24:01,950
IT IS TO REALLY, YOU KNOW, BUILD

674
00:24:01,950 --> 00:24:03,619
AND PREPARE A DATASET THAT COULD

675
00:24:03,619 --> 00:24:06,789
THEN GO ON AND BE USED FOR MODEL

676
00:24:06,789 --> 00:24:09,058
DEVELOPMENT AND THEN FINALLY FOR

677
00:24:09,058 --> 00:24:12,294
EVALUATION AND DEPLOYMENT.

678
00:24:12,294 --> 00:24:14,897
AND SO THIS SORT OF SQUARE UP 

679
00:24:14,897 --> 00:24:17,399
HERE IN THE UPPER PART OF THIS 

680
00:24:17,399 --> 00:24:20,202
LIFE CYCLE IS WHAT THE BRIDGE2AI

681
00:24:20,202 --> 00:24:22,204
CONSORTIUM IS REALLY FOCUSED ON.

682
00:24:22,204 --> 00:24:24,540
THE CONSORTIUM AND THE DATA 

683
00:24:24,540 --> 00:24:26,975
GENERATION PROJECTS ORIGINALLY 

684
00:24:26,975 --> 00:24:29,078
WHAT WERE BROKEN UP INTO SIX 

685
00:24:29,078 --> 00:24:32,915
DIFFERENT MODULES AND HAVE NOW 

686
00:24:32,915 --> 00:24:35,184
BEEN REORGANIZED INTO THESE 

687
00:24:35,184 --> 00:24:37,252
THREE PILLARS I TALKED ABOUT 

688
00:24:37,252 --> 00:24:37,486
BEFORE.

689
00:24:37,486 --> 00:24:39,621
IT IS SUPPOSED TO GIVE US NOT 

690
00:24:39,621 --> 00:24:42,057
ONLY THE DATA REPOSITORIES, BUT 

691
00:24:42,057 --> 00:24:43,959
ALSO THE BLUEPRINT OF HOW TO 

692
00:24:43,959 --> 00:24:45,794
GENERATE THESE TYPES OF DATASETS

693
00:24:45,794 --> 00:24:47,062
IN THE FUTURE.

694
00:24:47,062 --> 00:24:48,931
IN THE BRIDGE2AI PROGRAM, THERE 

695
00:24:48,931 --> 00:24:51,166
ARE FOUR DATA GENERATION 

696
00:24:51,166 --> 00:24:51,834
PROJECTS.

697
00:24:51,834 --> 00:24:54,636
THE FIRST THAT IS LISTED HERE IS

698
00:24:54,636 --> 00:24:59,575
THE CELL MAPS FOR AI OR THE 

699
00:24:59,575 --> 00:25:02,544
FUNCTIONAL GENOMICS GRANT 

700
00:25:02,544 --> 00:25:02,811
CHALLENGE.

701
00:25:02,811 --> 00:25:04,346
THIS IS BASIC SCIENCE FOCUSED 

702
00:25:04,346 --> 00:25:05,681
DATASET WHERE THEY ARE TRYING TO

703
00:25:05,681 --> 00:25:08,050
BUILD CELL MAPS OF EVERY CELL 

704
00:25:08,050 --> 00:25:09,852
TYPE IN THE HUMAN BODY.

705
00:25:09,852 --> 00:25:12,921
THE SECOND IS A VOICE DATA 

706
00:25:12,921 --> 00:25:14,823
INFORMATION PROJECT THAT IS 

707
00:25:14,823 --> 00:25:18,026
TRYING TO COLLECT THE DATASET TO

708
00:25:18,026 --> 00:25:20,629
STUDY HOW HUMAN SPOKEN SPEECH 

709
00:25:20,629 --> 00:25:23,499
COULD BE USED AS A BIOMARKER OF 

710
00:25:23,499 --> 00:25:23,732
HEALTH.

711
00:25:23,732 --> 00:25:25,701
IT HAS VERY DIFFERENT DOMAINS 

712
00:25:25,701 --> 00:25:28,871
AND DATA TYPES TIED WITH VOICE 

713
00:25:28,871 --> 00:25:30,372
RECORDINGS TO SEE IF MODELS 

714
00:25:30,372 --> 00:25:33,008
COULD LEARN THE STATE OF 

715
00:25:33,008 --> 00:25:35,410
SOMEBODY'S HEALTH BASED ON 

716
00:25:35,410 --> 00:25:36,745
SOMEBODY'S VOICE.

717
00:25:36,745 --> 00:25:39,148
THE THIRD IS OUR PROJECT KNOWN 

718
00:25:39,148 --> 00:25:42,084
AS SALUTOGENESIS OR AI-READI.

719
00:25:42,084 --> 00:25:43,418
I WILL GO INTO MUCH MORE DETAIL 

720
00:25:43,418 --> 00:25:45,420
ABOUT THAT DATASET IN JUST A 

721
00:25:45,420 --> 00:25:45,654
SECOND.

722
00:25:45,654 --> 00:25:48,590
THE LAST DATASET HERE IS KNOWN 

723
00:25:48,590 --> 00:25:53,996
AS THE CRITICAL CARE OR CHORUS 

724
00:25:53,996 --> 00:25:55,264
DATASET, MAINLY FOCUSED ON 

725
00:25:55,264 --> 00:25:58,467
PEOPLE IN THE ICU AND THE DATA 

726
00:25:58,467 --> 00:25:59,668
ASSOCIATED WITH THOSE PEOPLE TO 

727
00:25:59,668 --> 00:26:03,438
SEE IF WE CAN FIGURE OUT HOW TO 

728
00:26:03,438 --> 00:26:06,141
PREVENT MORTALITY IN THE ICU OR 

729
00:26:06,141 --> 00:26:07,543
READMISSION BACK TO THE 

730
00:26:07,543 --> 00:26:07,809
HOSPITAL.

731
00:26:07,809 --> 00:26:11,113
SO THESE ARE THE FOUR DATA 

732
00:26:11,113 --> 00:26:13,382
GENERATION PROJECTS PART OF THE 

733
00:26:13,382 --> 00:26:14,116
BRIDGE2AI PROGRAM.

734
00:26:14,116 --> 00:26:18,453
SO WHEN I SAY OURS, I'M 

735
00:26:18,453 --> 00:26:22,057
REFERRING TO CECILIA, MY WIFE, 

736
00:26:22,057 --> 00:26:27,229
THE CO-PI AND ME, AI READY ATLAS

737
00:26:27,229 --> 00:26:28,330
FOR DIABETES INSIGHTS.

738
00:26:28,330 --> 00:26:30,766
WE CAME UP WITH THIS IDEA AND 

739
00:26:30,766 --> 00:26:32,968
BLUEPRINT OF STUDYING 

740
00:26:32,968 --> 00:26:34,036
SALUTOGENESIS IN TYPE 2 

741
00:26:34,036 --> 00:26:34,303
DIABETES.

742
00:26:34,303 --> 00:26:36,572
SO THE FUNDAMENTAL GOAL OF OUR 

743
00:26:36,572 --> 00:26:38,807
DATA GENERATION PROJECT IS TO 

744
00:26:38,807 --> 00:26:40,542
CREATE A MULTIDIMENSIONAL, 

745
00:26:40,542 --> 00:26:42,744
ETHICALLY SOURCED DATASET IN 

746
00:26:42,744 --> 00:26:47,382
DIVERSE PEOPLE FOR STUDYING 

747
00:26:47,382 --> 00:26:48,317
SALUTOGENESIS IN TYPE 2 DIAB

748
00:26:48,317 --> 00:26:48,584
DIABETES.

749
00:26:48,584 --> 00:26:49,451
HOPEFULLY THIS AUDIENCE IS 

750
00:26:49,451 --> 00:26:50,786
FAMILIAR WITH THE TERM 

751
00:26:50,786 --> 00:26:51,119
SALUTOGENESIS.

752
00:26:51,119 --> 00:26:52,854
BUT IF YOU ARE NOT, IT CAN BE 

753
00:26:52,854 --> 00:26:55,791
THOUGHT OF AS THE REVERSE 

754
00:26:55,791 --> 00:27:00,662
PROCESS OF PATHOGENESIS.

755
00:27:00,662 --> 00:27:02,531
WE'VE SPENT -- REALLY THE MAIN 

756
00:27:02,531 --> 00:27:04,233
FOCUS OF A LOT OF BIOMEDICAL 

757
00:27:04,233 --> 00:27:05,734
RESEARCH FOR THE LAST CENTURY OR

758
00:27:05,734 --> 00:27:08,870
TWO HAS BEEN STUDYING THE 

759
00:27:08,870 --> 00:27:09,638
PATHWAYS OF DISEASE.

760
00:27:09,638 --> 00:27:11,940
HOW DOES SOMEONE GO FROM BEING 

761
00:27:11,940 --> 00:27:13,108
HEALTHY TO DISEASED?

762
00:27:13,108 --> 00:27:15,277
HOW DO WE SLOW IT DOWN OR 

763
00:27:15,277 --> 00:27:18,780
IDEALLY SORT OF STOP IT?

764
00:27:18,780 --> 00:27:22,117
WE SPENT VERY LITTLE TIME AND 

765
00:27:22,117 --> 00:27:23,218
ENERGY THINKING ABOUT THE 

766
00:27:23,218 --> 00:27:25,354
REVERSE PROCESS OF HOW DO WE 

767
00:27:25,354 --> 00:27:26,321
ACTUALLY RETURN SOMEBODY BACK TO

768
00:27:26,321 --> 00:27:28,724
A HEALTHY STATE?

769
00:27:28,724 --> 00:27:31,760
AND TYPE 2 DIABETES IS A DISEASE

770
00:27:31,760 --> 00:27:34,796
WHERE I SORT OF BELIEVE THAT A 

771
00:27:34,796 --> 00:27:37,666
SALUTOGENESIS TYPE PROCESS WILL 

772
00:27:37,666 --> 00:27:38,333
LIKELY EXIST.

773
00:27:38,333 --> 00:27:40,702
AND THAT'S THE REASON WHY WE 

774
00:27:40,702 --> 00:27:43,171
CHOSE IT AS A MODEL DISEASE FOR 

775
00:27:43,171 --> 00:27:47,643
THIS PARTICULAR PROJECT.

776
00:27:47,643 --> 00:27:51,179
OUR DATASET IS STRUCTURED IN A 

777
00:27:51,179 --> 00:27:52,281
TRIPLE BALANCED FASHION.

778
00:27:52,281 --> 00:27:55,517
AND I'LL EXPLAIN WHY THAT IS IN 

779
00:27:55,517 --> 00:27:57,252
JUST A SECOND.

780
00:27:57,252 --> 00:28:01,056
BUT WE REALLY WANTED TO HAVE AN 

781
00:28:01,056 --> 00:28:04,860
EVEN REPRESENTATION OF DIFFERENT

782
00:28:04,860 --> 00:28:07,029
RACE ETHNICITY GROUPS AS WELL AS

783
00:28:07,029 --> 00:28:08,163
BIOLOGICAL SEX.

784
00:28:08,163 --> 00:28:09,898
OF THE 4,000 PEOPLE WE ARE 

785
00:28:09,898 --> 00:28:11,066
TRYING TO COLLECT PEOPLE ON, 

786
00:28:11,066 --> 00:28:12,701
1,000 OF THE PEOPLE ARE SUPPOSED

787
00:28:12,701 --> 00:28:17,539
TO BE WHITE, 1,000 BLACK, 1,000 

788
00:28:17,539 --> 00:28:19,241
ASIAN AMERICAN AND 1,000 

789
00:28:19,241 --> 00:28:19,508
HISPANIC.

790
00:28:19,508 --> 00:28:22,244
WITHIN THAT WE WANTED TO SAMPLE 

791
00:28:22,244 --> 00:28:25,047
EVENLY ALONG THE DISEASE ACCESS 

792
00:28:25,047 --> 00:28:25,714
OF TYPE 2 DIABETES.

793
00:28:25,714 --> 00:28:28,283
AND THAT IS VERY IMPORTANT WHEN 

794
00:28:28,283 --> 00:28:29,851
YOU ARE DOING DIMENSION 

795
00:28:29,851 --> 00:28:31,153
REDUCTION TO MAKE SURE THAT ONE 

796
00:28:31,153 --> 00:28:39,127
OF THE GREATEST SOURCES OF 

797
00:28:39,127 --> 00:28:40,128
VARIABILITY IS THE DISEASE 

798
00:28:40,128 --> 00:28:42,431
ITSELF, THE DISEASE OF INTEREST.

799
00:28:42,431 --> 00:28:45,367
1,000 OF THESE PEOPLE WILL BE 

800
00:28:45,367 --> 00:28:47,402
NORMAL, 1,000 LIFESTYLE 

801
00:28:47,402 --> 00:28:50,972
CONTROLLED OR PREDIABETIC, 1,000

802
00:28:50,972 --> 00:28:52,641
CONTROLLED WITH ORAL MEDICATIONS

803
00:28:52,641 --> 00:28:57,012
AND 1,000 WILL BE DEPENDENT UPON

804
00:28:57,012 --> 00:28:57,245
INSULIN.

805
00:28:57,245 --> 00:28:59,081
WITHIN EACH ONE OF THESE BINS, 

806
00:28:59,081 --> 00:29:01,817
WE ARE HOPING TO ACHIEVE ONE TO 

807
00:29:01,817 --> 00:29:03,652
ONE BALANCING OF MALES AND 

808
00:29:03,652 --> 00:29:03,919
FEMALES.

809
00:29:03,919 --> 00:29:05,921
THIS IS VERY IMPORTANT BECAUSE 

810
00:29:05,921 --> 00:29:09,591
MOST BIOMEDICAL DATASETS ARE 

811
00:29:09,591 --> 00:29:10,459
PREDOMINANTLY FEMALE BECAUSE 

812
00:29:10,459 --> 00:29:12,627
FEMALES ARE MUCH MORE LIKELY TO 

813
00:29:12,627 --> 00:29:14,496
SEEK SORT OF CLINICAL CARE THAN 

814
00:29:14,496 --> 00:29:15,364
MALES ARE.

815
00:29:15,364 --> 00:29:17,899
SO IF YOU USE A ROUTINELY 

816
00:29:17,899 --> 00:29:21,536
COLLECTED DATASET, OFTEN IT IS 

817
00:29:21,536 --> 00:29:23,739
VERY DISEASED ENRICHED, OFTEN IT

818
00:29:23,739 --> 00:29:25,841
IS BIASED TOWARDS FEMALES, AND 

819
00:29:25,841 --> 00:29:27,843
OFTEN IT IS BIASED TOWARDS WHITE

820
00:29:27,843 --> 00:29:28,343
PEOPLE.

821
00:29:28,343 --> 00:29:30,512
THIS MAKES IT VERY CONCERNING 

822
00:29:30,512 --> 00:29:31,546
BECAUSE WHEN WE TRAIN DEEP 

823
00:29:31,546 --> 00:29:34,516
LEARNING MODELS AND WE GO AND 

824
00:29:34,516 --> 00:29:37,052
LET'S SAY WE HAVE AN AMAZING 

825
00:29:37,052 --> 00:29:40,322
DEEP LEARNING MODEL THAT CAN 

826
00:29:40,322 --> 00:29:42,657
DIAGNOSE DIABETES, IF WE TRAINED

827
00:29:42,657 --> 00:29:44,593
IT ON A ROUTINELY COLLECTED 

828
00:29:44,593 --> 00:29:46,795
DATASET, THERE IS A CHANCE WHEN 

829
00:29:46,795 --> 00:29:48,697
WE GO AND APPLY THIS MODEL TO 

830
00:29:48,697 --> 00:29:51,933
PEOPLE WHO ARE NON-WHITE, THAT 

831
00:29:51,933 --> 00:29:54,269
THE MODEL WILL MAKE A 

832
00:29:54,269 --> 00:29:56,004
CATASTROPHIC FAILURES OR 

833
00:29:56,004 --> 00:29:58,073
MISTAKES AND COULD LEAD TO, YOU 

834
00:29:58,073 --> 00:30:02,177
KNOW, HARM RATHER THAN GOOD.

835
00:30:02,177 --> 00:30:03,712
WE HAVE THREE DATA COLLECTION 

836
00:30:03,712 --> 00:30:06,014
SITES, ONE AFT UNIVERSITY OF 

837
00:30:06,014 --> 00:30:07,616
WASHINGTON, ONE AT UNIVERSITY OF

838
00:30:07,616 --> 00:30:11,420
ALABAMA AT BIRMINGHAM AND ONE AT

839
00:30:11,420 --> 00:30:14,222
UNIVERSITY OF CALIFORNIA AT SAN 

840
00:30:14,222 --> 00:30:14,423
DIEGO.

841
00:30:14,423 --> 00:30:16,658
I REALLY WANT TO SORT OF 

842
00:30:16,658 --> 00:30:18,160
EMPHASIZE THE POINT OF THIS 

843
00:30:18,160 --> 00:30:19,528
SLIDE AND THAT IS THAT, YOU 

844
00:30:19,528 --> 00:30:24,766
KNOW, IN OUR DATASET, WE ARE 

845
00:30:24,766 --> 00:30:26,401
REALLY COLLECTING DATA ELEMENTS 

846
00:30:26,401 --> 00:30:29,938
FROM HEAD TO TOE AND EVEN BEYOND

847
00:30:29,938 --> 00:30:30,705
THAT.

848
00:30:30,705 --> 00:30:33,942
AND SO, YOU KNOW, FROM -- 

849
00:30:33,942 --> 00:30:36,111
BECAUSE WE ARE SORT OF COME FROM

850
00:30:36,111 --> 00:30:41,683
THE OPHTHALMOLOGY FIELD, WE 

851
00:30:41,683 --> 00:30:44,085
SELECTED THAT ARE HIGHLY CUTTING

852
00:30:44,085 --> 00:30:47,222
EDGE AND RELEVANT TO TYPE 2 

853
00:30:47,222 --> 00:30:47,789
DIABETES.

854
00:30:47,789 --> 00:30:48,957
WE ARE COLLECTING THINGS FROM 

855
00:30:48,957 --> 00:30:52,127
PEOPLE'S HEARTS, EKG, MANY FORMS

856
00:30:52,127 --> 00:30:53,295
OF BLOODWORK THAT PROBE 

857
00:30:53,295 --> 00:30:55,497
EVERYTHING FROM KIDNEY FUNCTION 

858
00:30:55,497 --> 00:30:57,999
TO DIFFERENT MARKERS OF DIABETES

859
00:30:57,999 --> 00:31:00,001
AND HEART HEALTH.

860
00:31:00,001 --> 00:31:01,703
WE SENT PEOPLE HOME WITH THESE 

861
00:31:01,703 --> 00:31:03,138
DIFFERENT WEARABLE DEVICES AND 

862
00:31:03,138 --> 00:31:06,107
THEN WE ARE BANKING THESE 

863
00:31:06,107 --> 00:31:06,842
SAMPLES.

864
00:31:06,842 --> 00:31:08,343
THIS IS WHAT OUR DATA COLLECTION

865
00:31:08,343 --> 00:31:10,178
PROTOCOL LOOKS LIKE.

866
00:31:10,178 --> 00:31:12,547
IT IS A CROSS SECTIONAL DATASET.

867
00:31:12,547 --> 00:31:15,450
OF THE 4,000, WE ARE COLLECTING 

868
00:31:15,450 --> 00:31:18,286
ALL THIS INFORMATION ON THEM 

869
00:31:18,286 --> 00:31:18,920
ONCE.

870
00:31:18,920 --> 00:31:20,555
AND WHAT WE ARE ASKING PEOPLE TO

871
00:31:20,555 --> 00:31:22,357
DO BEFORE THEY COME IN IS A 

872
00:31:22,357 --> 00:31:23,692
WHOLE HOST OF DIFFERENT SURVEYS 

873
00:31:23,692 --> 00:31:26,061
THAT INCLUDES THINGS LIKE 

874
00:31:26,061 --> 00:31:29,064
DEPRESSION SCREENING, LOOKING AT

875
00:31:29,064 --> 00:31:30,265
DIFFERENT AREAS OF DIABETES 

876
00:31:30,265 --> 00:31:32,734
HEALTH, DIET SURVEY, SMOKING 

877
00:31:32,734 --> 00:31:34,803
HISTORY, AND SOCIAL DETERMINANTS

878
00:31:34,803 --> 00:31:36,404
OF HEALTH.

879
00:31:36,404 --> 00:31:39,774
AND DIFFERENT ACCESS TO HEALTH 

880
00:31:39,774 --> 00:31:41,076
CARE.

881
00:31:41,076 --> 00:31:43,245
AND SO THESE ARE ALL THE 

882
00:31:43,245 --> 00:31:44,246
DIFFERENT SURVEYS WE ASK PEOPLE 

883
00:31:44,246 --> 00:31:45,780
TO DO AT HOME BEFORE THEY COME 

884
00:31:45,780 --> 00:31:46,114
IN.

885
00:31:46,114 --> 00:31:48,650
WHEN THEY ACTUALLY COME IN FOR 

886
00:31:48,650 --> 00:31:50,051
THE CLINICAL SITE VISIT IT IS 

887
00:31:50,051 --> 00:31:52,487
ABOUT A FOUR-HOUR VISIT WHERE 

888
00:31:52,487 --> 00:31:55,123
THEY COME IN AND WE DO A WHOLE 

889
00:31:55,123 --> 00:31:58,326
HOST OF DIFFERENT THINGS WITH 

890
00:31:58,326 --> 00:31:58,793
THEM.

891
00:31:58,793 --> 00:32:03,665
ONE IS, OF COURSE, RETINA 

892
00:32:03,665 --> 00:32:06,401
IMAGING, OCT AND THE PHOTOS I 

893
00:32:06,401 --> 00:32:09,070
SHOWED YOU EXAMPLES OF BEFORE.

894
00:32:09,070 --> 00:32:11,907
WE OBVIOUSLY COLLECT THINGS LIKE

895
00:32:11,907 --> 00:32:16,111
HEIGHT, WEIGHT, WAIST AND HIP 

896
00:32:16,111 --> 00:32:16,745
CIRCUMFERENCE, BLOOD PRESSURE 

897
00:32:16,745 --> 00:32:17,712
AND HEART RATE.

898
00:32:17,712 --> 00:32:19,548
WE DO SOME VISION TESTING.

899
00:32:19,548 --> 00:32:22,217
WE COLLECT ALL THEIR MEDICATION 

900
00:32:22,217 --> 00:32:22,584
HISTORY.

901
00:32:22,584 --> 00:32:24,319
WE ALSO ASK THEM FOR PERMISSION 

902
00:32:24,319 --> 00:32:26,555
TO PULL THEIR DRIVING RECORD TO 

903
00:32:26,555 --> 00:32:28,390
UNDERSTAND HOW THEIR DRIVING 

904
00:32:28,390 --> 00:32:29,758
BEHAVIOR IS AND WHETHER THEY 

905
00:32:29,758 --> 00:32:31,660
HAVE ANY ACCIDENTS THAT ARE 

906
00:32:31,660 --> 00:32:33,461
LISTED IN THEIR NAME.

907
00:32:33,461 --> 00:32:36,064
WE TAKE OFF THEIR SHOES AND WE 

908
00:32:36,064 --> 00:32:43,672
DO A MONOFILAMENT TEST TO TEST 

909
00:32:43,672 --> 00:32:49,611
FOR PERIPHERAL NEUROTHOPY, WE 

910
00:32:49,611 --> 00:32:53,615
PERFORM A TEST CALLED MOCA, A 

911
00:32:53,615 --> 00:32:56,418
WHOLE HOST OF BLOOD TESTS AND 

912
00:32:56,418 --> 00:33:01,056
TAKE THEIR BIOSPECIMENS AND MAIL

913
00:33:01,056 --> 00:33:03,725
TO THE UNIVERSITY OF ALABAMA, 

914
00:33:03,725 --> 00:33:05,927
BIRMINGHAM WHERE THEY ARE CRYO 

915
00:33:05,927 --> 00:33:09,965
FROZEN TO BE USED AS ASSAYS IN 

916
00:33:09,965 --> 00:33:10,699
THE FUTURE.

917
00:33:10,699 --> 00:33:15,103
FOR ABOUT 10 DAYS AT HOME, THEY 

918
00:33:15,103 --> 00:33:21,076
ARE WEARING A DEX CON G6 GLUE 

919
00:33:21,076 --> 00:33:25,013
KOES  -- GLUCOSE MONITORING 

920
00:33:25,013 --> 00:33:25,513
DEVICE.

921
00:33:25,513 --> 00:33:29,784
WE ASK THEM TO WEAR A GARMIN 

922
00:33:29,784 --> 00:33:30,452
FINANCE TRACKER.

923
00:33:30,452 --> 00:33:33,054
AND FINALLY WE BUILT AN 

924
00:33:33,054 --> 00:33:34,489
ENVIRONMENTAL SENSOR.

925
00:33:34,489 --> 00:33:37,425
THAT CAN MEASURE THINGS LIKE 

926
00:33:37,425 --> 00:33:39,294
TEMPERATURE, HUMIDITY, BUT ALSO 

927
00:33:39,294 --> 00:33:40,428
PARTICULATE MATTER INSIDE OF 

928
00:33:40,428 --> 00:33:45,066
THEIR HOMES AND THINGS LIKE 

929
00:33:45,066 --> 00:33:46,001
ORGANIC COMPOUNDS.

930
00:33:46,001 --> 00:33:52,774
IT HAS A SPECTROGRAM THAT CAN 

931
00:33:52,774 --> 00:33:54,643
MEASURE LIGHT IN PEOPLE'S HOME 

932
00:33:54,643 --> 00:33:56,811
SO WE CAN TELL AS THEY TURN 

933
00:33:56,811 --> 00:33:58,947
LIGHTS ON AND OFF AS THE HOUSE 

934
00:33:58,947 --> 00:34:03,885
GETS DARK, WE CAN LOOK AT THE 

935
00:34:03,885 --> 00:34:06,921
SPECTRUM OF LIGHT AND TELL 

936
00:34:06,921 --> 00:34:17,432
WHETHER THEY HAVE FLUORESCENT 

937
00:34:20,435 --> 00:34:21,603
LIGHTBULBS.

938
00:34:21,603 --> 00:34:23,705
THE OTHER THING THAT IS REALLY 

939
00:34:23,705 --> 00:34:25,907
IMPORTANT IS THERE IS A LOT OF 

940
00:34:25,907 --> 00:34:27,509
TALK WHETHER GAS STOVES ARE SAFE

941
00:34:27,509 --> 00:34:28,176
OR NOT.

942
00:34:28,176 --> 00:34:30,211
ONE OF THE BIG PROBLEMS WHEN YOU

943
00:34:30,211 --> 00:34:32,080
ARE LOOKING AT ENVIRONMENTAL 

944
00:34:32,080 --> 00:34:34,416
RESEARCH HAS BEEN THAT PEOPLE 

945
00:34:34,416 --> 00:34:36,818
HAVE MAINLY BEEN MAPPING FIVE 

946
00:34:36,818 --> 00:34:39,854
DIGIT ZIP CODES TO ATMOSPHERIC 

947
00:34:39,854 --> 00:34:43,425
DATA AND PRESUMES THAT IS THE 

948
00:34:43,425 --> 00:34:45,627
AIR THAT PEOPLE ARE BEING 

949
00:34:45,627 --> 00:34:47,295
EXPOSED TO.

950
00:34:47,295 --> 00:34:49,664
MOST PEOPLE ARE SPENDING, YOU 

951
00:34:49,664 --> 00:34:51,232
KNOW, THEIR WINTER MONTHS INSIDE

952
00:34:51,232 --> 00:34:52,834
OF THEIR HOMES AND PROBABLY AT 

953
00:34:52,834 --> 00:34:54,903
LEAST 50% OF THEIR LIFE WHEN 

954
00:34:54,903 --> 00:34:56,805
THEY ARE ASLEEP, YOU KNOW, IS 

955
00:34:56,805 --> 00:34:58,206
BEING SPENT INSIDE OF THEIR 

956
00:34:58,206 --> 00:34:59,774
HOMES, BREATHING THE AIR INSIDE 

957
00:34:59,774 --> 00:35:00,942
OF THEIR HOMES.

958
00:35:00,942 --> 00:35:03,411
SO IT IS REALLY IMPORTANT, IN MY

959
00:35:03,411 --> 00:35:05,814
OPINION, TO DO THIS RESEARCH BY 

960
00:35:05,814 --> 00:35:07,215
MEASURING THE AIR SAMPLE INSIDE 

961
00:35:07,215 --> 00:35:08,316
OF THEIR HOMES TO UNDERSTAND 

962
00:35:08,316 --> 00:35:11,720
WHAT KIND OF AIR THEY ARE 

963
00:35:11,720 --> 00:35:12,987
BREATHING.

964
00:35:12,987 --> 00:35:15,490
SO I WANT TO SPEND A LITTLE BIT 

965
00:35:15,490 --> 00:35:20,228
OF TIME HERE TALKING ABOUT THIS 

966
00:35:20,228 --> 00:35:21,596
DIMENSION REDUCTION TECHNIQUE OR

967
00:35:21,596 --> 00:35:22,797
MANIFOLD LEARNING I WAS TALKING 

968
00:35:22,797 --> 00:35:24,132
ABOUT IN THE PREVIOUS SECTION.

969
00:35:24,132 --> 00:35:26,201
SO WE'RE GOING TO GO THROUGH 

970
00:35:26,201 --> 00:35:27,969
SORT OF TWO MODEL PATIENTS.

971
00:35:27,969 --> 00:35:30,171
YOU KNOW, LET'S PRETEND THAT WE 

972
00:35:30,171 --> 00:35:33,074
HAVE A 41-YEAR-OLD MALE WHO HAS 

973
00:35:33,074 --> 00:35:34,876
REALLY NO PAST MEDICAL HISTORY, 

974
00:35:34,876 --> 00:35:37,045
BUT HAS HAD POOR DIETARY HABITS 

975
00:35:37,045 --> 00:35:38,980
FOR MOST OF HIS LIFE AND, YOU 

976
00:35:38,980 --> 00:35:41,049
KNOW, QUITE FRANKLY, THIS MIGHT 

977
00:35:41,049 --> 00:35:41,750
BE ME.

978
00:35:41,750 --> 00:35:44,085
AND, YOU KNOW, A COUPLE OF YEARS

979
00:35:44,085 --> 00:35:47,956
LATER THEY END UP HAVING A MORE 

980
00:35:47,956 --> 00:35:49,991
SEDENTARY LIFESTYLE AND THEN BY 

981
00:35:49,991 --> 00:35:51,960
THE TIME THEY TURN 47, THEY 

982
00:35:51,960 --> 00:35:54,929
NEVER REALLY WENT FOR A PHYSICAL

983
00:35:54,929 --> 00:35:55,163
CHECKUP.

984
00:35:55,163 --> 00:35:58,433
AND THEN BY THE AGE OF 49, THEY 

985
00:35:58,433 --> 00:36:00,235
START DEVELOPING COMPLICATIONS 

986
00:36:00,235 --> 00:36:02,937
FROM UNTREATED DIABETES.

987
00:36:02,937 --> 00:36:05,540
THESE COLOR DOTS SORT OF GO FROM

988
00:36:05,540 --> 00:36:06,641
HEALTHY DOWN TO DISEASE AND 

989
00:36:06,641 --> 00:36:08,610
REPRESENTS SORT OF THE BODY 

990
00:36:08,610 --> 00:36:10,411
STATE, THE WHOLE PERSON HEALTH 

991
00:36:10,411 --> 00:36:14,382
STATE OF THIS PARTICULAR MODEL 

992
00:36:14,382 --> 00:36:14,883
PATIENT.

993
00:36:14,883 --> 00:36:17,051
AND WHAT'S BEING PLOTTED HERE ON

994
00:36:17,051 --> 00:36:22,423
THE X AXIS IS TIME, SINCE SORT 

995
00:36:22,423 --> 00:36:24,626
OF THE FIRST CLINICAL VISIT OF 

996
00:36:24,626 --> 00:36:24,993
THIS PERSON.

997
00:36:24,993 --> 00:36:27,695
SO IF YOU TAKE A DIFFERENT 

998
00:36:27,695 --> 00:36:30,965
PERSON AND LET'S SAY THIS OTHER 

999
00:36:30,965 --> 00:36:32,200
MODEL PARTICIPANT IS 28 YEARS 

1000
00:36:32,200 --> 00:36:35,069
OLD, MALE, NO PAST MEDICAL 

1001
00:36:35,069 --> 00:36:37,071
HISTORY, AROUND 31 THEY STOPPED 

1002
00:36:37,071 --> 00:36:38,506
EXERCISING REGULARLY, BECAME 

1003
00:36:38,506 --> 00:36:42,010
MORE SEDENTARY, PICKED UP SOME 

1004
00:36:42,010 --> 00:36:44,012
BAD DIETARY HABITS BUT FINALLY 

1005
00:36:44,012 --> 00:36:46,981
GOES TO GET A PHYSICAL AND GETS 

1006
00:36:46,981 --> 00:36:47,916
DIAGNOSED WITH EARLY DIABETES 

1007
00:36:47,916 --> 00:36:50,285
AND REALLY THAT ENDS UP BEING A 

1008
00:36:50,285 --> 00:36:51,886
WAKEUP CALL.

1009
00:36:51,886 --> 00:36:54,355
HE STARTS SEEING A HEALTH COACH,

1010
00:36:54,355 --> 00:36:55,456
EXERCISING REGULARLY, WATCHING 

1011
00:36:55,456 --> 00:36:59,194
WHAT HE HEATS AND HE HOPEFULLY 

1012
00:36:59,194 --> 00:37:00,228
STARTS REVERSING HIS HEALTH 

1013
00:37:00,228 --> 00:37:02,363
STATE BACK TO A HEALTHY STATE.

1014
00:37:02,363 --> 00:37:04,833
ONE OF THE POINTS I WANT TO MAKE

1015
00:37:04,833 --> 00:37:09,070
HERE IS THAT THE BODY STATE IS 

1016
00:37:09,070 --> 00:37:11,005
VERY UNLIKELY THAT THEY ARE 

1017
00:37:11,005 --> 00:37:12,407
GOING TO BE IDENTICAL WHEN THEY 

1018
00:37:12,407 --> 00:37:15,977
ARE SORT OF GOING DOWN THIS WAY 

1019
00:37:15,977 --> 00:37:19,214
ON THIS GRAPH VERSUS STARTING TO

1020
00:37:19,214 --> 00:37:21,316
TRY TO TURN THE TIDE AND LIVE A 

1021
00:37:21,316 --> 00:37:22,550
MORE HEALTHY LIFESTYLE AFTER 

1022
00:37:22,550 --> 00:37:23,918
BEING AT A LOWER POINT ON THIS 

1023
00:37:23,918 --> 00:37:25,053
GRAPH.

1024
00:37:25,053 --> 00:37:27,222
SO IT IS VERY UNLIKELY THAT 

1025
00:37:27,222 --> 00:37:28,389
THESE TWO STATES ARE EXACTLY THE

1026
00:37:28,389 --> 00:37:31,759
SAME.

1027
00:37:31,759 --> 00:37:35,163
SO IF YOU TAKE THESE TWO MODEL 

1028
00:37:35,163 --> 00:37:36,397
PARTICIPANTS AND THEN YOU PUT 

1029
00:37:36,397 --> 00:37:38,433
THEM ON THE SAME GRAPH AND YOU 

1030
00:37:38,433 --> 00:37:40,869
USE AGE, ONE WAS YOUNGER, ONE 

1031
00:37:40,869 --> 00:37:43,304
WAS MUCH OLDER, THEY WILL END UP

1032
00:37:43,304 --> 00:37:44,973
SORT OF LOOKING LIKE THIS.

1033
00:37:44,973 --> 00:37:46,608
WHERE YOU CAN'T REALLY MAKE 

1034
00:37:46,608 --> 00:37:48,376
HEADS OR TAILS OUT OF THEM.

1035
00:37:48,376 --> 00:37:50,778
BUT INSTEAD, WHAT WE TRY TO DO 

1036
00:37:50,778 --> 00:37:53,548
IS WE TRY TO ALIGN THE BODY 

1037
00:37:53,548 --> 00:37:54,115
STATES TOGETHER.

1038
00:37:54,115 --> 00:37:56,651
WE CAN SORT OF COMBINE THESE 

1039
00:37:56,651 --> 00:37:58,887
GRAPHS TOGETHER AND IF WE SAMPLE

1040
00:37:58,887 --> 00:38:01,522
LOTS AND LOTS OF PEOPLE, WE WILL

1041
00:38:01,522 --> 00:38:03,224
HOPEFULLY FIND SORT OF THIS 

1042
00:38:03,224 --> 00:38:04,525
GRAPH WHERE THERE ARE PEOPLE 

1043
00:38:04,525 --> 00:38:06,628
SORT OF GOING THROUGH SIMILAR 

1044
00:38:06,628 --> 00:38:08,429
BODY STATES AS THEY TRANSITION 

1045
00:38:08,429 --> 00:38:10,932
FROM HEALTHY TO DISEASED AND 

1046
00:38:10,932 --> 00:38:12,367
SIMILAR BODY STATES AS THEY GO 

1047
00:38:12,367 --> 00:38:16,137
FROM DISEASED BACK TO HEALTHY.

1048
00:38:16,137 --> 00:38:19,974
AND THESE -- THIS X AXIS WENT 

1049
00:38:19,974 --> 00:38:21,976
FROM TIME TO AGE TO NOW PSEUDO 

1050
00:38:21,976 --> 00:38:24,012
TIME, WHICH PSEUDO TIME IS NOT A

1051
00:38:24,012 --> 00:38:26,714
REAL TIME INTERVAL, BUT IT IS 

1052
00:38:26,714 --> 00:38:28,750
BEING COLLAPSED BY SIMILAR 

1053
00:38:28,750 --> 00:38:32,120
POSITIONS OF END STATES OF 

1054
00:38:32,120 --> 00:38:32,353
HEALTH.

1055
00:38:32,353 --> 00:38:35,390
AND SO THAT'S SORT OF THE GRAND 

1056
00:38:35,390 --> 00:38:36,858
IDEA BEHIND OUR DATASET.

1057
00:38:36,858 --> 00:38:41,863
WHAT WE ARE HOPING TO DO IS 

1058
00:38:41,863 --> 00:38:44,632
COLLECT THIS VERY LARGE 

1059
00:38:44,632 --> 00:38:49,637
MULTIMODAL DATASET EXPANDS FROM 

1060
00:38:49,637 --> 00:38:52,106
HEAD TO TOE AND EXAMINES THE 

1061
00:38:52,106 --> 00:38:53,875
ENVIRONMENT AROUND THEM AND 

1062
00:38:53,875 --> 00:38:56,678
BUILD A DATASET TO BE USED FOR 

1063
00:38:56,678 --> 00:38:59,280
DIMENSIONAL REDUCTION TO BUILD 

1064
00:38:59,280 --> 00:39:01,082
ONE DIMENSIONAL MANIFOLDS IN 

1065
00:39:01,082 --> 00:39:03,418
THIS LOWER DIMENSIONAL SPACE 

1066
00:39:03,418 --> 00:39:06,754
WHERE ONE AXIS IS PSEUDOTIME AND

1067
00:39:06,754 --> 00:39:09,791
THE OTHER AXIS IS HEALTHY AND 

1068
00:39:09,791 --> 00:39:10,058
DISEASED.

1069
00:39:10,058 --> 00:39:14,929
BY DOING SO WE CAN STUDY 

1070
00:39:14,929 --> 00:39:16,164
DIFFERENT THINGS LIKE 

1071
00:39:16,164 --> 00:39:17,699
PATHOGENESIS AND SALUTOGENESIS.

1072
00:39:17,699 --> 00:39:19,367
THAT IS THE FUNDAMENTAL IDEA 

1073
00:39:19,367 --> 00:39:21,202
BEHIND OUR DATASET AND THE GRAND

1074
00:39:21,202 --> 00:39:22,870
CHALLENGE THAT WE ARE BUILDING.

1075
00:39:22,870 --> 00:39:26,307
NOW ONE THING THAT I'M SORT OF 

1076
00:39:26,307 --> 00:39:29,277
EXCITED TO SHOW YOU IS THAT IN 

1077
00:39:29,277 --> 00:39:31,346
DECEMBER OF LAST YEAR, WE 

1078
00:39:31,346 --> 00:39:33,247
COMPLETED OUR PILOT AND WE HAD 

1079
00:39:33,247 --> 00:39:36,651
ABOUT 204 PARTICIPANTS IN THERE.

1080
00:39:36,651 --> 00:39:43,958
AND THIS DATASET HAD ABOUT 202, 

1081
00:39:43,958 --> 00:39:48,329
EKGs, 28,000 LAB MEASUREMENTS.

1082
00:39:48,329 --> 00:39:49,864
FROM THE WEARABLE DEVICES WE 

1083
00:39:49,864 --> 00:39:51,733
HAVE MILLIONS OF DATA POINTS OF 

1084
00:39:51,733 --> 00:39:56,838
WHAT THEIR GLUCOSE WAS AND HEART

1085
00:39:56,838 --> 00:39:58,973
RATE AND OXYGEN SATURATION, 

1086
00:39:58,973 --> 00:40:00,174
STRESS LEVEL AND HOW MANY HOURS 

1087
00:40:00,174 --> 00:40:03,211
OF SLEEP THEY WERE GETTING.

1088
00:40:03,211 --> 00:40:07,281
WE HAVE 33 MILLION SAMPLES, 

1089
00:40:07,281 --> 00:40:09,584
RECORDINGS OF DIFFERENT AIR 

1090
00:40:09,584 --> 00:40:12,954
POLLUTION LEVELS AS WELL AS 

1091
00:40:12,954 --> 00:40:20,428
LIGHT SPECTROMETRY MEETINGS AND 

1092
00:40:20,428 --> 00:40:21,896
RETINAL IMAGING.

1093
00:40:21,896 --> 00:40:26,768
THIS IS 21,000 FILES TAKING UP 

1094
00:40:26,768 --> 00:40:31,139
311 GIGABYTES OF DATA, 2500 

1095
00:40:31,139 --> 00:40:32,240
BIOSPECIMENS FROM THIS COHORT 

1096
00:40:32,240 --> 00:40:35,777
HAVE BEEN BANKED.

1097
00:40:35,777 --> 00:40:37,245
WHAT I'M REALLY EXCITED TO SHARE

1098
00:40:37,245 --> 00:40:40,681
IS A COUPLE OF MONTHS AGO WE 

1099
00:40:40,681 --> 00:40:41,783
PUBLISHED THIS DATASET.

1100
00:40:41,783 --> 00:40:43,051
IT IS OPENLY AVAILABLE FOR 

1101
00:40:43,051 --> 00:40:45,420
PEOPLE TO GO TO OUR WEBSITE AND 

1102
00:40:45,420 --> 00:40:46,921
REGISTER AND DOWNLOAD THIS 

1103
00:40:46,921 --> 00:40:49,924
DATASET TO YOUR COMPUTERS TO DO 

1104
00:40:49,924 --> 00:40:50,525
ANALYSIS WITH.

1105
00:40:50,525 --> 00:40:52,493
WE HAVE A WHOLE HOST OF 

1106
00:40:52,493 --> 00:40:55,730
DIFFERENT MECHANISMS THAT WE PUT

1107
00:40:55,730 --> 00:40:58,933
IN PLACE TO SHARE IT IN THE MOST

1108
00:40:58,933 --> 00:41:00,201
SAFE AND RESPONSIBILITY WAY 

1109
00:41:00,201 --> 00:41:00,468
POSSIBLE.

1110
00:41:00,468 --> 00:41:02,770
BUT WE ALSO WANTED TO GIVE THE 

1111
00:41:02,770 --> 00:41:04,672
ABILITY FOR PEOPLE TO ACTUALLY 

1112
00:41:04,672 --> 00:41:06,374
DOWNLOAD THE DATASET TO THEIR 

1113
00:41:06,374 --> 00:41:08,709
COMPUTERS, USE THE TOOLS AND 

1114
00:41:08,709 --> 00:41:12,713
ANALYSIS PACKAGES THEY ARE 

1115
00:41:12,713 --> 00:41:14,415
FAMILIAR WITH AND DISCOVER NEW 

1116
00:41:14,415 --> 00:41:14,649
THINGS.

1117
00:41:14,649 --> 00:41:16,684
THIS IS JUST FROM OUR PILOT.

1118
00:41:16,684 --> 00:41:19,487
IT HAS 204 PARTICIPANTS, BUT WE 

1119
00:41:19,487 --> 00:41:20,888
ARE AIMING FOR 4,000 PEOPLE.

1120
00:41:20,888 --> 00:41:22,924
SO THE FULL DATASET SHOULD BE 

1121
00:41:22,924 --> 00:41:23,991
ABOUT 20 TIMES.

1122
00:41:23,991 --> 00:41:25,760
ALL THESE NUMBERS SHOULD BE 

1123
00:41:25,760 --> 00:41:27,195
MULTIPLIED BY 20 BY THE TIME WE 

1124
00:41:27,195 --> 00:41:28,663
GET TO THE END OF IT.

1125
00:41:28,663 --> 00:41:30,765
WE DO HOPE TO SHARE IT IN THE 

1126
00:41:30,765 --> 00:41:31,599
SAME WAY.

1127
00:41:31,599 --> 00:41:33,434
EVERY YEAR WE PLAN TO RELEASE 

1128
00:41:33,434 --> 00:41:35,169
ALL THE DATA THAT WE HAVE.

1129
00:41:35,169 --> 00:41:38,806
OUR NEXT DATA RELEASE IS SLATED 

1130
00:41:38,806 --> 00:41:42,743
FOR NOVEMBER.

1131
00:41:42,743 --> 00:41:45,179
WE ARE ALREADY PAST 800 SOME 

1132
00:41:45,179 --> 00:41:47,415
PARTICIPANTS IN OUR DATASET NOW 

1133
00:41:47,415 --> 00:41:48,816
AND SO BY THE TIME WE GET TO 

1134
00:41:48,816 --> 00:41:51,786
NOVEMBER WE ARE HOPING TO 

1135
00:41:51,786 --> 00:41:54,455
RELEASE 1,000 PEOPLE WORTH OF 

1136
00:41:54,455 --> 00:42:00,761
DATA FOR PEOPLE TO USE FOR 

1137
00:42:00,761 --> 00:42:01,028
ANALYSIS.

1138
00:42:01,028 --> 00:42:02,930
WE HAVE A NUMBER OF DIFFERENT 

1139
00:42:02,930 --> 00:42:03,197
PARTNERS.

1140
00:42:03,197 --> 00:42:05,032
MANY OF THESE COMPANIES WERE 

1141
00:42:05,032 --> 00:42:06,934
GRACIOUS ENOUGH TO EITHER LEND 

1142
00:42:06,934 --> 00:42:09,804
US THEIR DEVICE OR PROVIDE 

1143
00:42:09,804 --> 00:42:11,305
DISCOUNT PRICING.

1144
00:42:11,305 --> 00:42:13,374
ONE PARTNER IN PARTICULAR, 

1145
00:42:13,374 --> 00:42:16,577
MICROSOFT WAS ABLE TO KIND OF 

1146
00:42:16,577 --> 00:42:18,479
GIVE US FREE CLOUD CREDITS TO 

1147
00:42:18,479 --> 00:42:21,048
USE TO BUILD AND SHARE THIS 

1148
00:42:21,048 --> 00:42:22,917
DATASET WITH THE WORLD.

1149
00:42:22,917 --> 00:42:27,255
I JUST WANT TO END BY SORT OF 

1150
00:42:27,255 --> 00:42:28,356
ACKNOWLEDGING THE MANY, MANY 

1151
00:42:28,356 --> 00:42:29,457
PEOPLE THAT ARE BEING SUPPORTED 

1152
00:42:29,457 --> 00:42:31,826
BY THIS PROJECT.

1153
00:42:31,826 --> 00:42:33,928
THERE'S MANY ELEMENTS OF THIS 

1154
00:42:33,928 --> 00:42:35,696
THAT -- OF THIS PROGRAM, OUR 

1155
00:42:35,696 --> 00:42:37,865
PROJECT IN PARTICULAR, THAT I 

1156
00:42:37,865 --> 00:42:41,736
DIDN'T HAVE TIME TO SORT OF GO 

1157
00:42:41,736 --> 00:42:41,969
INTO.

1158
00:42:41,969 --> 00:42:45,273
FOR EXAMPLE, SALLY BAXTER AT 

1159
00:42:45,273 --> 00:42:47,208
UCSD IS LEADING AN INTERNSHIP 

1160
00:42:47,208 --> 00:42:49,510
PROGRAM WHERE WE ARE TAKING 

1161
00:42:49,510 --> 00:42:51,379
PEOPLE FROM DIVERSE BACKGROUNDS 

1162
00:42:51,379 --> 00:42:54,849
AND BRINGING THEM INTO THE 

1163
00:42:54,849 --> 00:42:56,317
DIFFERENT INVESTIGATORS' LABS IN

1164
00:42:56,317 --> 00:42:58,553
THIS PROJECT AND HELPING THEM 

1165
00:42:58,553 --> 00:43:01,055
LEARN HOW TO USE AI AND TRAIN 

1166
00:43:01,055 --> 00:43:03,691
DEEP LEARNING MODELS AND LEARN 

1167
00:43:03,691 --> 00:43:05,026
MACHINE LEARNING.

1168
00:43:05,026 --> 00:43:07,361
WE HAVE A WHOLE GROUP OF PEOPLE 

1169
00:43:07,361 --> 00:43:09,063
THAT IS TRYING TO STANDARDIZE 

1170
00:43:09,063 --> 00:43:11,766
DATA SO IT IS VERY EASY TO USE 

1171
00:43:11,766 --> 00:43:14,769
DATA ONCE YOU DOWNLOADED IT AND 

1172
00:43:14,769 --> 00:43:17,138
YOU DON'T NEED TO SPEND AN INORD

1173
00:43:17,138 --> 00:43:18,806
NANT AMOUNT OF TIME ORGANIZING 

1174
00:43:18,806 --> 00:43:21,075
THE DATA BEFORE YOU START USING 

1175
00:43:21,075 --> 00:43:23,311
STATISTICAL MODELS OR TRAINING 

1176
00:43:23,311 --> 00:43:25,813
DEEP LEARNING MODELS.

1177
00:43:25,813 --> 00:43:33,354
WE HAVE, YOU KNOW, A GROUP AT 

1178
00:43:33,354 --> 00:43:36,157
JOE -- AND NATIVE BIODATA IS 

1179
00:43:36,157 --> 00:43:37,558
ENGAGING WITH AN AMERICAN INDIAN

1180
00:43:37,558 --> 00:43:39,961
TRIBE TO SEE IF THEY MIGHT BE 

1181
00:43:39,961 --> 00:43:42,396
INTERESTED IN REPLICATING OUR 

1182
00:43:42,396 --> 00:43:45,066
PROTOCOL TO COLLECT DATA INSIDE 

1183
00:43:45,066 --> 00:43:47,335
OF THEIR TRIBE.

1184
00:43:47,335 --> 00:43:49,971
SO THEY MAY, IF THEY, SORT OF 

1185
00:43:49,971 --> 00:43:52,640
THE IDEA BEHIND THAT IS IF THEY 

1186
00:43:52,640 --> 00:43:54,041
FOLLOW OUR COLLECTION PROTOCOL 

1187
00:43:54,041 --> 00:43:57,111
AND USE THE EXACT SAME STEPS 

1188
00:43:57,111 --> 00:43:59,880
THEN THAT DATA WOULD BECOME 

1189
00:43:59,880 --> 00:44:01,415
HARMONIZABLE WITH THE LARGER 

1190
00:44:01,415 --> 00:44:01,649
DATASET.

1191
00:44:01,649 --> 00:44:04,085
THERE IS ANOTHER GROUP OF FOLKS 

1192
00:44:04,085 --> 00:44:05,620
LIKE SARAH SINGER AT STANFORD 

1193
00:44:05,620 --> 00:44:08,122
THAT IS TRYING TO UNDERSTAND HOW

1194
00:44:08,122 --> 00:44:10,024
DIFFERENT MULTIDISCIPLINARY 

1195
00:44:10,024 --> 00:44:11,325
TEAMS CAN INTERACT TOGETHER IN 

1196
00:44:11,325 --> 00:44:11,759
TEAM SCIENCE.

1197
00:44:11,759 --> 00:44:13,995
THAT IS JUST A FLAVOR OF THE 

1198
00:44:13,995 --> 00:44:16,831
VERY LARGE GROUP OF PEOPLE THAT 

1199
00:44:16,831 --> 00:44:17,765
ARE BEHIND THIS PROJECT TRYING 

1200
00:44:17,765 --> 00:44:21,068
TO MAKE IT A SUCCESS.

1201
00:44:21,068 --> 00:44:24,105
THIS IS THE QR CODE TO OUR 

1202
00:44:24,105 --> 00:44:26,073
WEBSITE WHERE YOU CAN, YOU KNOW,

1203
00:44:26,073 --> 00:44:28,776
VIEW THE DOCUMENTATION AND 

1204
00:44:28,776 --> 00:44:30,177
ACCESS THE DATASET.

1205
00:44:30,177 --> 00:44:32,013
WHEN YOU GO TO ACCESS THE 

1206
00:44:32,013 --> 00:44:34,315
DATASET IT WILL TALK YOU THROUGH

1207
00:44:34,315 --> 00:44:36,284
MANY DIFFERENT STEPS BEFORE YOU 

1208
00:44:36,284 --> 00:44:37,918
CAN DOWNLOAD THIS DATASET.

1209
00:44:37,918 --> 00:44:39,787
ONE OF THOSE ACTUALLY INVOLVES 

1210
00:44:39,787 --> 00:44:41,922
SORT OF WATERMARKING THE DATASET

1211
00:44:41,922 --> 00:44:44,992
SO THAT DATASET YOU DOWNLOAD IS 

1212
00:44:44,992 --> 00:44:47,094
DIGITALLY TRACEABLE BACK TO YOU 

1213
00:44:47,094 --> 00:44:49,664
SO IF YOU SORT OF VIOLATE OUR 

1214
00:44:49,664 --> 00:44:52,333
TERMS AND YOU RESHARE THAT 

1215
00:44:52,333 --> 00:44:54,035
DATASET, WE WOULD KNOW SORT OF 

1216
00:44:54,035 --> 00:44:56,537
WHO YOU ARE AND BE ABLE TO TRACK

1217
00:44:56,537 --> 00:44:57,071
YOU DOWN.

1218
00:44:57,071 --> 00:44:59,273
SO THERE ARE SORT OF ALL SORTS 

1219
00:44:59,273 --> 00:45:00,708
OF THINGS THAT WE'VE DONE IN 

1220
00:45:00,708 --> 00:45:04,111
THIS PROJECT TO SORT OF TRY AND 

1221
00:45:04,111 --> 00:45:06,147
BUILD A BLUEPRINT FOR HOW TO 

1222
00:45:06,147 --> 00:45:09,450
COLLECT SUCH A DATASET IN THE 

1223
00:45:09,450 --> 00:45:12,486
FUTURE AND RELEASE IT SAFELY TO 

1224
00:45:12,486 --> 00:45:13,087
THE WORLD.

1225
00:45:13,087 --> 00:45:14,922
SO I'M HOPING AND GUESSING THAT 

1226
00:45:14,922 --> 00:45:17,058
THERE WILL BE LOTS OF QUESTIONS 

1227
00:45:17,058 --> 00:45:19,827
THAT WE CAN SPEND SOME TIME 

1228
00:45:19,827 --> 00:45:22,163
GOING OVER AND I GUESS I'LL STOP

1229
00:45:22,163 --> 00:45:23,764
SHARING NOW AND BE HAPPY TO TAKE

1230
00:45:23,764 --> 00:45:28,269
THOSE QUESTIONS.

1231
00:45:28,269 --> 00:45:30,438
>> WELL, THANK YOU SO
MUCH, DR. 

1232
00:45:30,438 --> 00:45:31,706
LEE, THAT WAS JUST RICH WITH 

1233
00:45:31,706 --> 00:45:32,306
MANY THINGS.

1234
00:45:32,306 --> 00:45:35,009
A LOT OF DATA AND A LOT OF 

1235
00:45:35,009 --> 00:45:35,943
REALLY INTERESTING FUTURE 

1236
00:45:35,943 --> 00:45:36,711
DIRECTIONS.

1237
00:45:36,711 --> 00:45:37,978
SO IT SOUNDS LIKE THINGS ARE 

1238
00:45:37,978 --> 00:45:39,246
GOING TO GO.

1239
00:45:39,246 --> 00:45:41,315
I KNOW THAT DAVID, I THINK YOU 

1240
00:45:41,315 --> 00:45:45,986
HAVE SOME QUESTIONS TO KICK US 

1241
00:45:45,986 --> 00:45:47,321
OFF WITH.

1242
00:45:47,321 --> 00:45:49,090
SO DAVID, WHAT IS YOUR FIRST 

1243
00:45:49,090 --> 00:45:49,990
QUESTION FOR DR. LEE?

1244
00:45:49,990 --> 00:45:51,726
>> FIRST OF ALL, AARON,
THANK 

1245
00:45:51,726 --> 00:45:53,194
YOU SO MUCH.

1246
00:45:53,194 --> 00:45:55,296
YOU HAVE SHOWN US THE FUTURE OF 

1247
00:45:55,296 --> 00:45:55,963
HEALTH CARE RESEARCH AND HEALTH 

1248
00:45:55,963 --> 00:45:56,497
CARE NOW.

1249
00:45:56,497 --> 00:45:58,666
YOU ARE SHOWING US THE WAY HOW 

1250
00:45:58,666 --> 00:46:03,070
WE CAN REALLY USE RESEARCH TO 

1251
00:46:03,070 --> 00:46:04,872
BENEFIT PEOPLE, RESTORE THEIR 

1252
00:46:04,872 --> 00:46:05,940
HEALTH, MAINTAIN THEIR HEALTH 

1253
00:46:05,940 --> 00:46:07,174
AND CURE DISEASE.

1254
00:46:07,174 --> 00:46:11,612
SO I THANK YOU FOR WHAT YOU'RE 

1255
00:46:11,612 --> 00:46:11,879
DOING.

1256
00:46:11,879 --> 00:46:14,281
IT ALSO POINTS TO THIS IDEA, 

1257
00:46:14,281 --> 00:46:16,250
THIS LEARNING HEALTH CARE SYSTEM

1258
00:46:16,250 --> 00:46:18,052
IT IS INTERTWINED PATIENT CARE 

1259
00:46:18,052 --> 00:46:19,253
OF RESEARCH THAT FIT TOGETHER IN

1260
00:46:19,253 --> 00:46:22,089
WAYS OF PROMOTING HEALTH AND 

1261
00:46:22,089 --> 00:46:22,590
HEALTH CARE.

1262
00:46:22,590 --> 00:46:24,392
SO THANK YOU FOR ALL YOU ARE 

1263
00:46:24,392 --> 00:46:24,725
DOING.

1264
00:46:24,725 --> 00:46:26,694
I GUESS JUST A COUPLE OF 

1265
00:46:26,694 --> 00:46:27,595
QUESTIONS AROUND BIG DATA AND 

1266
00:46:27,595 --> 00:46:30,331
THE FACT YOU RELEASED YOUR FIRST

1267
00:46:30,331 --> 00:46:31,665
WAVE OF DATA.

1268
00:46:31,665 --> 00:46:34,735
CAN YOU SPEAK TO THE ISSUES 

1269
00:46:34,735 --> 00:46:36,837
AROUND FAIR, FINDABLE, 

1270
00:46:36,837 --> 00:46:38,672
ACCESSIBLE, INTEROPERABLE AND 

1271
00:46:38,672 --> 00:46:39,206
REUSABLE?

1272
00:46:39,206 --> 00:46:42,643
WE HAVE ALL OF US PROGRAM AT NIH

1273
00:46:42,643 --> 00:46:44,879
THE UK DATA BANK, THE WORK YOU 

1274
00:46:44,879 --> 00:46:47,782
ARE DOING AND OTHERS THROUGH 

1275
00:46:47,782 --> 00:46:48,349
BRIDGE2AI.

1276
00:46:48,349 --> 00:46:50,418
ARE WE ALL WORKING -- WE ARE ALL

1277
00:46:50,418 --> 00:46:52,219
WORKING TOWARDS THE SAME GOAL.

1278
00:46:52,219 --> 00:46:54,922
ARE THERE WAYS TO HARNESS ALL 

1279
00:46:54,922 --> 00:46:56,390
THESE DATA, MAKE THEM 

1280
00:46:56,390 --> 00:46:57,258
INTEROPERABLE, SHAREABLE?

1281
00:46:57,258 --> 00:46:59,326
YOUR THOUGHTS ON WHERE WE ARE 

1282
00:46:59,326 --> 00:47:00,795
GOING WITH BIG DATA IN GENERAL 

1283
00:47:00,795 --> 00:47:02,563
AROUND PRINCIPLES OF FAIR?

1284
00:47:02,563 --> 00:47:03,264
>> YEAH.

1285
00:47:03,264 --> 00:47:07,535
SO WE, AND WHEN I SAY WE, I MEAN

1286
00:47:07,535 --> 00:47:09,170
THE FIELD OF MEDICINE WRIT 

1287
00:47:09,170 --> 00:47:09,537
LARGE.

1288
00:47:09,537 --> 00:47:11,539
I THINK IT IS REALLY IMPORTANT 

1289
00:47:11,539 --> 00:47:13,774
THAT SORT OF THE ROADS OF 

1290
00:47:13,774 --> 00:47:17,445
PROGRESS LEAD TO A PLACE WHERE 

1291
00:47:17,445 --> 00:47:19,647
DATA IS ULTIMATELY HARMONIZABLE 

1292
00:47:19,647 --> 00:47:22,116
ACROSS ALL OF THESE DIFFERENT 

1293
00:47:22,116 --> 00:47:22,383
EFFORTS.

1294
00:47:22,383 --> 00:47:23,984
AND SO, YOU KNOW, WITHIN THE 

1295
00:47:23,984 --> 00:47:25,052
BRIDGE2AI PROGRAM, THERE IS A 

1296
00:47:25,052 --> 00:47:27,988
LOT OF WORK THAT IS BEING DONE 

1297
00:47:27,988 --> 00:47:31,292
AT THE CENTER LEVEL AS WELL AS 

1298
00:47:31,292 --> 00:47:34,562
BETWEEN DIFFERENT DATA 

1299
00:47:34,562 --> 00:47:35,596
GENERATION PROJECTS TO MAKE SURE

1300
00:47:35,596 --> 00:47:38,365
WHEN, LIKE, ONE PROJECT IS USING

1301
00:47:38,365 --> 00:47:39,700
A PARTICULAR DATA TYPE THAT IT 

1302
00:47:39,700 --> 00:47:43,204
IS FORMATTED IN THE SAME WAY AS 

1303
00:47:43,204 --> 00:47:44,238
WHEN ANOTHER DATA GENERATION 

1304
00:47:44,238 --> 00:47:45,706
PROJECT IS USING THAT DATA TYPE.

1305
00:47:45,706 --> 00:47:48,609
SO THAT ULTIMATELY IN THE END, 

1306
00:47:48,609 --> 00:47:50,978
YOU COULD EVEN JUST COMBINE ALL 

1307
00:47:50,978 --> 00:47:51,946
THE FILES TOGETHER AND ANALYZE 

1308
00:47:51,946 --> 00:47:55,049
THEM IN THE SAME WAY.

1309
00:47:55,049 --> 00:47:56,917
THAT ALSO SORT OF LEADS TO SOME 

1310
00:47:56,917 --> 00:47:59,353
OF THE EFFORTS THAT WE HAVE, 

1311
00:47:59,353 --> 00:48:02,223
THAT WE ARE A PART OF AT THE 

1312
00:48:02,223 --> 00:48:03,190
COMMON FUND LEVEL.

1313
00:48:03,190 --> 00:48:05,226
THERE IS SOMETHING KNOWN AS THE 

1314
00:48:05,226 --> 00:48:08,429
COMMON FUND DATA ECOSYSTEM.

1315
00:48:08,429 --> 00:48:09,430
WE ARE, YOU KNOW, WE'VE SPENT 

1316
00:48:09,430 --> 00:48:11,198
THE FIRST COUPLE OF YEARS OF THE

1317
00:48:11,198 --> 00:48:13,701
PROJECT GETTING OUR DATASET 

1318
00:48:13,701 --> 00:48:15,803
READY AND STANDARDIZED AND 

1319
00:48:15,803 --> 00:48:17,438
HARMONIZED AND NOW IT IS SORT OF

1320
00:48:17,438 --> 00:48:19,540
THE NEXT STEP TO BUILD THAT 

1321
00:48:19,540 --> 00:48:20,941
HARMONY AT A HIGHER LEVEL.

1322
00:48:20,941 --> 00:48:23,811
SO THAT INVOLVES WORKING WITH 

1323
00:48:23,811 --> 00:48:26,847
THE CFDE TO TRY AND MAKE SURE 

1324
00:48:26,847 --> 00:48:28,816
THAT THE DATA ELEMENTS THAT WE 

1325
00:48:28,816 --> 00:48:30,885
HAVE MATCH THE META DATA THEY 

1326
00:48:30,885 --> 00:48:32,520
HAVE IN THEIR SORT OF ECOSYSTEM.

1327
00:48:32,520 --> 00:48:35,656
AND THEN THERE ARE TALKS WITH 

1328
00:48:35,656 --> 00:48:37,658
THE ALL OF US PROGRAM AS WELL TO

1329
00:48:37,658 --> 00:48:40,761
MAKE SURE THAT OUR DATA IS 

1330
00:48:40,761 --> 00:48:42,363
INTERCHANGEABLE WITH, YOU KNOW, 

1331
00:48:42,363 --> 00:48:44,198
SOME OF THE DATA THAT IS BEING 

1332
00:48:44,198 --> 00:48:46,934
COLLECTED IN THE ALL OF US 

1333
00:48:46,934 --> 00:48:47,601
PROGRAM.

1334
00:48:47,601 --> 00:48:50,337
SO I HOPE THAT PART OF THIS BLUE

1335
00:48:50,337 --> 00:48:52,172
PRINT THAT PEOPLE TAKE HOME WHEN

1336
00:48:52,172 --> 00:48:53,807
THEY ARE BUILDING DATASETS OR 

1337
00:48:53,807 --> 00:48:55,643
THINKING ABOUT SHARING DATASETS 

1338
00:48:55,643 --> 00:48:57,578
INCLUDES THIS IDEA THAT IT NEEDS

1339
00:48:57,578 --> 00:49:00,080
TO BE, THE DATA NEEDS TO BE IN 

1340
00:49:00,080 --> 00:49:01,582
SOME SORT OF STANDARD THAT OTHER

1341
00:49:01,582 --> 00:49:03,350
PEOPLE ARE VERY FAMILIAR WITH 

1342
00:49:03,350 --> 00:49:06,620
THAT IS EASY TO USE AND IS OPEN.

1343
00:49:06,620 --> 00:49:10,991
SO THAT IT CAN SORT OF BE 

1344
00:49:10,991 --> 00:49:12,660
HARMONIZED AT A MUCH, MUCH 

1345
00:49:12,660 --> 00:49:13,427
HIGHER LEVEL THAN A SINGLE 

1346
00:49:13,427 --> 00:49:16,130
DATASET.

1347
00:49:16,130 --> 00:49:17,264
>> I'M GOING TO JUMP IN
HERE 

1348
00:49:17,264 --> 00:49:19,567
JUST FOR A SECOND BEFORE WE COME

1349
00:49:19,567 --> 00:49:21,402
BACK TO YOU DAVID FOR A FOLLOW 

1350
00:49:21,402 --> 00:49:22,002
UP QUESTION.

1351
00:49:22,002 --> 00:49:24,204
I'M GOING TO REMIND OUR 

1352
00:49:24,204 --> 00:49:26,173
VIDEOCAST AUDIENCE THAT THE 

1353
00:49:26,173 --> 00:49:27,741
FEEDBACK FORM LINK IS RIGHT 

1354
00:49:27,741 --> 00:49:29,043
BELOW THE SCREEN YOU ARE 

1355
00:49:29,043 --> 00:49:29,276
VIEWING.

1356
00:49:29,276 --> 00:49:32,613
YOU ARE WELCOME TO SEND IN 

1357
00:49:32,613 --> 00:49:33,013
QUESTIONS.

1358
00:49:33,013 --> 00:49:35,082
AND I'M HOPING, YOU KNOW, SOME 

1359
00:49:35,082 --> 00:49:40,321
FOLKS ARE FURIOUSLY TYPING AND 

1360
00:49:40,321 --> 00:49:46,226
THE NCCIH EVENTS ON E-MAIL IS 

1361
00:49:46,226 --> 00:49:48,062
AVAILABLE TO SEND E-MAILS THERE 

1362
00:49:48,062 --> 00:49:48,362
SEPARATELY.

1363
00:49:48,362 --> 00:49:50,464
MAKE SURE TO AVAIL YOURSELF OF 

1364
00:49:50,464 --> 00:49:50,664
THAT.

1365
00:49:50,664 --> 00:49:53,834
SO DAVID, DID YOU HAVE A FOLLOW 

1366
00:49:53,834 --> 00:49:55,736
UP QUESTION?

1367
00:49:55,736 --> 00:49:58,639
>> I HAVE SOME OTHER
QUESTIONS, 

1368
00:49:58,639 --> 00:50:00,441
BUT I WANT TO OPEN TO THE 

1369
00:50:00,441 --> 00:50:00,741
AUDIENCE.

1370
00:50:00,741 --> 00:50:01,842
I WILL COME BACK TO AARON IF 

1371
00:50:01,842 --> 00:50:03,811
TIME ALLOWS.

1372
00:50:03,811 --> 00:50:06,080
THANK YOU SO MUCH FOR ENGAGING 

1373
00:50:06,080 --> 00:50:07,948
ABOUT A FUTURISTIC TALK ABOUT 

1374
00:50:07,948 --> 00:50:11,318
WHERE WE CAN GO WITH BIG DATA 

1375
00:50:11,318 --> 00:50:17,124
AND MACHINE LEARNING.

1376
00:50:17,124 --> 00:50:22,529
>> HELENE, WHAT QUESTIONS
DO YOU

1377
00:50:22,529 --> 00:50:22,730
HAVE?

1378
00:50:22,730 --> 00:50:26,634
>> YOU GAVE US A KIND OF
AN 

1379
00:50:26,634 --> 00:50:29,436
OPENING VIEW OF WHAT IT IS LIKE 

1380
00:50:29,436 --> 00:50:31,538
TO ANALYZE THESE BIG DATASETS, 

1381
00:50:31,538 --> 00:50:33,040
VERY, VERY EXCITING.

1382
00:50:33,040 --> 00:50:35,075
I HAVE A QUESTION ABOUT THE 

1383
00:50:35,075 --> 00:50:37,845
EXAMPLE THAT YOU USED AT THE 

1384
00:50:37,845 --> 00:50:41,048
BEGINNING ABOUT HOW YOU WERE 

1385
00:50:41,048 --> 00:50:46,353
ABLE TO GET THIS VERY HIGHLY 

1386
00:50:46,353 --> 00:50:48,589
ACCURATE ESSENTIALLY DIAGNOSIS 

1387
00:50:48,589 --> 00:50:50,491
OF MACULAR DEGENERATION BASED ON

1388
00:50:50,491 --> 00:51:01,001
A SIMPLE SET OF TOMS THERE -- 

1389
00:51:04,071 --> 00:51:04,304
TOMOGRAPHY.

1390
00:51:04,304 --> 00:51:05,039
BASED ON HUMANS.

1391
00:51:05,039 --> 00:51:07,508
>> RIGHT.

1392
00:51:07,508 --> 00:51:08,342
>> PRESUMABLY SOME OTHER
TYPES 

1393
00:51:08,342 --> 00:51:09,443
OF DATA.

1394
00:51:09,443 --> 00:51:12,379
SO MY QUESTION THEN IS WHAT IF 

1395
00:51:12,379 --> 00:51:13,881
YOUR DEEP LEARNING MODEL 

1396
00:51:13,881 --> 00:51:17,051
DISCOVERS SOMETHING COMPLETELY 

1397
00:51:17,051 --> 00:51:19,687
NEW THAT WE DON'T KNOW?

1398
00:51:19,687 --> 00:51:23,257
WHAT IS THE PROCESS BY WHICH YOU

1399
00:51:23,257 --> 00:51:26,760
WOULD BE ABLE TO BRING THIS BACK

1400
00:51:26,760 --> 00:51:28,796
TO ESSENTIALLY TO TEST IT?

1401
00:51:28,796 --> 00:51:32,466
AND THIS WOULD APPLY, I SUPPOSE,

1402
00:51:32,466 --> 00:51:34,501
TO THIS SALUTOGENESIS BECAUSE WE

1403
00:51:34,501 --> 00:51:36,403
DON'T HAVE OTHER WAYS, RIGHT, TO

1404
00:51:36,403 --> 00:51:38,472
UNDERSTAND THE MECHANISMS OF 

1405
00:51:38,472 --> 00:51:39,206
SALUTOGENESIS.

1406
00:51:39,206 --> 00:51:41,341
SO SAYING YOU FIND SOME VERY, 

1407
00:51:41,341 --> 00:51:43,844
VERY INTERESTING NEW MANIFOLD 

1408
00:51:43,844 --> 00:51:45,279
THAT SEEMS TO DESCRIBE 

1409
00:51:45,279 --> 00:51:46,146
INDIVIDUALS WHO ARE GOING BACK 

1410
00:51:46,146 --> 00:51:48,148
TO HEALTH, WHAT WOULD BE YOUR 

1411
00:51:48,148 --> 00:51:51,385
NEXT STEP TO VERIFY THAT THAT IS

1412
00:51:51,385 --> 00:51:52,252
ACTUALLY THE CASE?

1413
00:51:52,252 --> 00:51:52,586
>> YEAH.

1414
00:51:52,586 --> 00:51:54,788
SO THAT IS A REALLY GREAT 

1415
00:51:54,788 --> 00:51:57,458
QUESTION AND AN INSIGHTFUL ONE.

1416
00:51:57,458 --> 00:51:59,626
SO LET ME SORT OF BOIL DOWN THAT

1417
00:51:59,626 --> 00:52:01,128
QUESTION INTO A COUPLE OF 

1418
00:52:01,128 --> 00:52:04,698
DIFFERENT CONCEPTS.

1419
00:52:04,698 --> 00:52:07,234
ONE IS THAT TO YOUR POINT, IF WE

1420
00:52:07,234 --> 00:52:09,069
TRAIN A DEEP LEARNING MODEL 

1421
00:52:09,069 --> 00:52:11,472
USING HUMAN LABELS, SO LIKE IN 

1422
00:52:11,472 --> 00:52:13,741
THIS PARTICULAR EXAMPLE IT WAS 

1423
00:52:13,741 --> 00:52:16,076
NORMAL VERSUS MACULAR 

1424
00:52:16,076 --> 00:52:16,410
DEGENERATION.

1425
00:52:16,410 --> 00:52:18,879
THE MODEL IS GOING TO 

1426
00:52:18,879 --> 00:52:20,347
RECAPITULATE OUR CLINICAL 

1427
00:52:20,347 --> 00:52:21,215
UNDERSTANDING, RIGHT, BECAUSE 

1428
00:52:21,215 --> 00:52:22,816
THAT IS LITERALLY WHAT IT IS 

1429
00:52:22,816 --> 00:52:25,319
BEING TRAINED TO DO.

1430
00:52:25,319 --> 00:52:28,021
IT IS GIVEN THE SAME LABELS 

1431
00:52:28,021 --> 00:52:28,989
DERIVED FROM A HUMAN 

1432
00:52:28,989 --> 00:52:29,823
UNDERSTANDING OF DISEASE.

1433
00:52:29,823 --> 00:52:32,626
SO THERE IS A ELEMENT TO WHAT 

1434
00:52:32,626 --> 00:52:33,961
THOSE MODELS WILL DISCOVER.

1435
00:52:33,961 --> 00:52:35,963
AND THE WAY THAT WE VERIFIED 

1436
00:52:35,963 --> 00:52:36,864
THAT IN THAT PARTICULAR EXAMPLE 

1437
00:52:36,864 --> 00:52:39,399
IS USING THIS SORT OF 

1438
00:52:39,399 --> 00:52:40,667
OCCLUSION-BASED TECHNIQUE WHERE 

1439
00:52:40,667 --> 00:52:42,136
WE HID DIFFERENT PARTS OF THE 

1440
00:52:42,136 --> 00:52:44,104
IMAGE AND LOOKED TO SEE WHAT 

1441
00:52:44,104 --> 00:52:46,406
HAPPENED TO THE MODEL CONFIDENCE

1442
00:52:46,406 --> 00:52:48,942
IN CALLING THAT MACULAR 

1443
00:52:48,942 --> 00:52:50,611
DEGENERATION AND WE COULD BUILD 

1444
00:52:50,611 --> 00:52:52,746
SORT OF THAT HEAT MAP.

1445
00:52:52,746 --> 00:52:54,314
THAT TECHNIQUE AND THAT 

1446
00:52:54,314 --> 00:52:55,849
VERIFICATION METHOD WORKS WHEN 

1447
00:52:55,849 --> 00:52:57,985
YOU ARE WITHIN THE CONTEXT OF 

1448
00:52:57,985 --> 00:53:00,120
SORT OF KNOWN DISEASES AND KNOWN

1449
00:53:00,120 --> 00:53:01,054
MODEL BEHAVIORS.

1450
00:53:01,054 --> 00:53:02,422
SO THERE IS A WHOLE CLASS OF 

1451
00:53:02,422 --> 00:53:03,791
DEEP LEARNING WHERE THE GOAL 

1452
00:53:03,791 --> 00:53:07,427
THERE IS TO REALLY TRY AND DO 

1453
00:53:07,427 --> 00:53:10,397
WHAT HUMAN BEINGS DO, BUT 

1454
00:53:10,397 --> 00:53:10,998
AUTOMATE IT.

1455
00:53:10,998 --> 00:53:12,933
SO THAT IS SORT OF THE WAY THAT 

1456
00:53:12,933 --> 00:53:15,302
YOU WOULD VERIFY THAT, IS BY 

1457
00:53:15,302 --> 00:53:17,070
CHECKING IT AGAINST HUMAN 

1458
00:53:17,070 --> 00:53:17,337
BEHAVIOR.

1459
00:53:17,337 --> 00:53:21,041
NOW, THERE IS ANOTHER WAY TO USE

1460
00:53:21,041 --> 00:53:23,010
DEEP LEARNING METHODS AND THAT 

1461
00:53:23,010 --> 00:53:24,344
IS FOR DISCOVERY.

1462
00:53:24,344 --> 00:53:26,046
THAT IS WHERE WE ARE TRYING TO 

1463
00:53:26,046 --> 00:53:27,481
PUSH THE BOUNDS OF HUMAN 

1464
00:53:27,481 --> 00:53:27,748
KNOWLEDGE.

1465
00:53:27,748 --> 00:53:30,551
ONE OF THE WAYS WE CAN DO THAT 

1466
00:53:30,551 --> 00:53:32,519
IS BY USING LABELS OR WHAT WE 

1467
00:53:32,519 --> 00:53:35,522
USE TO TRAIN THESE MODELS, THAT 

1468
00:53:35,522 --> 00:53:40,661
INFORMATION NEEDS TO COME FROM 

1469
00:53:40,661 --> 00:53:42,496
SOME SORT OF DATA ELEMENTS THAT 

1470
00:53:42,496 --> 00:53:46,667
IS SORT OF IRREFUTABLE AND IS 

1471
00:53:46,667 --> 00:53:49,136
MUCH MORE TRUE THAN SORT OF 

1472
00:53:49,136 --> 00:53:49,937
HUMAN UNDERSTANDING.

1473
00:53:49,937 --> 00:53:51,405
AND SO LET ME SORT OF GIVE YOU 

1474
00:53:51,405 --> 00:53:53,340
AN EXAMPLE OF THAT.

1475
00:53:53,340 --> 00:53:57,044
SO IMAGINE YOU GOT AN EKG 

1476
00:53:57,044 --> 00:53:58,245
TRACING OF SOMEBODY AND YOU 

1477
00:53:58,245 --> 00:54:00,647
KNEW, YOU HAD THE DATA THAT 

1478
00:54:00,647 --> 00:54:04,418
WITHIN 24 HOURS THAT PERSON HAD 

1479
00:54:04,418 --> 00:54:07,688
A CARDIAC EVENT.

1480
00:54:07,688 --> 00:54:09,289
YOU -- HUMAN BEINGS MIGHT NOT BE

1481
00:54:09,289 --> 00:54:12,226
ABLE TO LOOK AT AN EKG AND BE 

1482
00:54:12,226 --> 00:54:14,595
ABLE TO PREDICT WHETHER THAT 

1483
00:54:14,595 --> 00:54:17,064
PERSON IS GOING TO HAVE A 

1484
00:54:17,064 --> 00:54:19,366
CARDIAC EVENT IN 24 TO 48 HOURS.

1485
00:54:19,366 --> 00:54:21,068
THERE MIGHT BE SOME SUBTLE 

1486
00:54:21,068 --> 00:54:21,535
CLUES.

1487
00:54:21,535 --> 00:54:22,836
THERE MAY NOT BE A FUNDAMENTAL 

1488
00:54:22,836 --> 00:54:25,439
SORT OF WAY TO ANALYZE THE EKGs 

1489
00:54:25,439 --> 00:54:30,577
WHERE WE CAN ACHIEVE A 98% 

1490
00:54:30,577 --> 00:54:32,913
SENSITIVITY AND SPECIFICITY 

1491
00:54:32,913 --> 00:54:34,281
AROUND THAT END POINT.

1492
00:54:34,281 --> 00:54:36,083
YOU CAN USE THAT AS THE GROUND 

1493
00:54:36,083 --> 00:54:37,651
TRUTH TO TRAIN THESE MODELS.

1494
00:54:37,651 --> 00:54:39,820
AND THE MODELS WILL LEARN 

1495
00:54:39,820 --> 00:54:41,054
DIRECTLY FROM THE DATA AND TRY 

1496
00:54:41,054 --> 00:54:42,656
TO LEARN TO DO THAT TASK.

1497
00:54:42,656 --> 00:54:45,525
IT MAY BE ABLE TO DO THINGS THAT

1498
00:54:45,525 --> 00:54:46,860
HUMAN BEINGS ARE NOT ABLE TO DO 

1499
00:54:46,860 --> 00:54:48,328
AND THEN LATER GO BACK AND TEACH

1500
00:54:48,328 --> 00:54:51,565
YOU THIS PART OF THE EKG IS WHAT

1501
00:54:51,565 --> 00:54:55,702
I USED TO BE ABLE TO MAKE THIS 

1502
00:54:55,702 --> 00:54:55,969
DIAGNOSIS.

1503
00:54:55,969 --> 00:54:58,872
AND SO THAT'S A CLASS OF SORT OF

1504
00:54:58,872 --> 00:55:02,542
DEEP LEARNING METHODS AND 

1505
00:55:02,542 --> 00:55:04,011
STRATEGIES YOU CAN USE TO 

1506
00:55:04,011 --> 00:55:05,812
DISCOVER NEW THINGS ABOUT HEALTH

1507
00:55:05,812 --> 00:55:08,715
AND DISEASE PROCESSES.

1508
00:55:08,715 --> 00:55:11,385
IF WE WERE -- IF THIS DATASET IS

1509
00:55:11,385 --> 00:55:12,653
INDEED SUCCESSFUL AND IT DOES 

1510
00:55:12,653 --> 00:55:14,922
LEAD TO SOME SORT OF BIOMARKER 

1511
00:55:14,922 --> 00:55:17,124
THAT IS PREDICTIVE OF 

1512
00:55:17,124 --> 00:55:18,625
SALUTOGENESIS, THE WAY YOU WOULD

1513
00:55:18,625 --> 00:55:21,528
GO AN VERIFY THOSE SORT OF NEW 

1514
00:55:21,528 --> 00:55:24,031
NOVEL BIOMARKERS WHERE THEY GO 

1515
00:55:24,031 --> 00:55:25,699
BEYOND HUMAN UNDERSTANDING IS TO

1516
00:55:25,699 --> 00:55:29,069
DO SORT OF A PROSPECTIVE 

1517
00:55:29,069 --> 00:55:29,369
EVALUATION.

1518
00:55:29,369 --> 00:55:31,038
WHERE YOU TAKE THIS MODEL AND 

1519
00:55:31,038 --> 00:55:34,775
YOU DEPLOY IT.

1520
00:55:34,775 --> 00:55:37,110
ALL THE PARTICIPANTS THAT ARE 

1521
00:55:37,110 --> 00:55:38,812
COMING IN THROUGH THE DOOR 

1522
00:55:38,812 --> 00:55:41,048
EITHER AT A HEALTH CLINIC OR 

1523
00:55:41,048 --> 00:55:42,649
HOSPITAL SYSTEM AND IN THE 

1524
00:55:42,649 --> 00:55:44,484
BACKGROUND YOU ARE RUNNING THIS 

1525
00:55:44,484 --> 00:55:47,321
MODEL TO SEE IF, INDEED, THERE 

1526
00:55:47,321 --> 00:55:50,857
ARE PEOPLE THAT HAVE THIS 

1527
00:55:50,857 --> 00:55:52,893
BIOMARKER THAT IS PREDICTIVE OF 

1528
00:55:52,893 --> 00:55:54,594
SOMEBODY TURNING BACK TO HEALTHY

1529
00:55:54,594 --> 00:55:54,795
STATE.

1530
00:55:54,795 --> 00:55:57,064
THEN YOU FOLLOW THEM INTO THE 

1531
00:55:57,064 --> 00:55:58,231
FUTURE AND YOU SEE WHAT HAPPENS 

1532
00:55:58,231 --> 00:56:00,434
TO THEM OR YOU CALL THEM BACK 

1533
00:56:00,434 --> 00:56:02,402
AND IDENTIFY THEM AND DO A 

1534
00:56:02,402 --> 00:56:06,239
CLINICAL STUDY WHERE YOU COLLECT

1535
00:56:06,239 --> 00:56:07,007
RELEVANT VARIABLES.

1536
00:56:07,007 --> 00:56:08,942
THEN YOU DO A VALIDATION STUDY 

1537
00:56:08,942 --> 00:56:11,945
TO SHOW, INDEED, IN THIS GROUP 

1538
00:56:11,945 --> 00:56:15,382
OF 50% OF THE COHORT POSITIVE 

1539
00:56:15,382 --> 00:56:19,953
FOR THIS BIOMARKER, THIS WAS THE

1540
00:56:19,953 --> 00:56:20,554
SPECIFICITY OF THOSE PEOPLE 

1541
00:56:20,554 --> 00:56:23,023
GOING ON TO A HEALTHY STATE AND 

1542
00:56:23,023 --> 00:56:26,360
THOSE PEOPLE BIOMARKER NEGATIVE 

1543
00:56:26,360 --> 00:56:29,062
WENT ON TO CONTINUE DOWN THE 

1544
00:56:29,062 --> 00:56:29,663
PATHOGENESIS PATHWAY.

1545
00:56:29,663 --> 00:56:32,165
YOU WOULD HAVE TO DO A STUDY 

1546
00:56:32,165 --> 00:56:35,335
LIKE THAT AT THE END OF THE DAY 

1547
00:56:35,335 --> 00:56:39,039
TO CONFIRM THAT THESE BIOMARKERS

1548
00:56:39,039 --> 00:56:42,275
ARE, INDEED, USEFUL AND 

1549
00:56:42,275 --> 00:56:44,578
CORRELATE WITH A MEANINGFUL 

1550
00:56:44,578 --> 00:56:45,078
CLINICAL OUTCOME.

1551
00:56:45,078 --> 00:56:46,113
>> THANK YOU.

1552
00:56:46,113 --> 00:56:48,215
IN THIS CASE, THE GROUND TRUTH 

1553
00:56:48,215 --> 00:56:49,516
WOULD END UP BEING TIME.

1554
00:56:49,516 --> 00:56:50,951
>> YES.

1555
00:56:50,951 --> 00:56:52,319
>> AS OPPOSED TO
PSEUDOTIME.

1556
00:56:52,319 --> 00:56:56,156
>> RIGHT, RIGHT, RIGHT.

1557
00:56:56,156 --> 00:56:56,456
ABSOLUTELY.

1558
00:56:56,456 --> 00:56:58,325
SO ONE OF THE THINGS THAT I WANT

1559
00:56:58,325 --> 00:57:00,761
TO SORT OF EMPHASIZE IS THAT 

1560
00:57:00,761 --> 00:57:02,629
BECAUSE WE ARE WORKING IN THE 

1561
00:57:02,629 --> 00:57:03,997
SPACE WHERE WE DON'T KNOW WHAT 

1562
00:57:03,997 --> 00:57:05,532
WE ARE LOOKING FOR, IT IS 

1563
00:57:05,532 --> 00:57:08,168
IMPORTANT THAT THIS DATASET BE 

1564
00:57:08,168 --> 00:57:10,871
DESIGNED FOR DISCOVERY AND FOR 

1565
00:57:10,871 --> 00:57:12,039
UNDERSTANDING DISEASE MECHANISMS

1566
00:57:12,039 --> 00:57:16,843
AND UNDERSTANDING NEW WAYS OF 

1567
00:57:16,843 --> 00:57:18,378
SALUTOGENESIS AND PATHOGENESIS.

1568
00:57:18,378 --> 00:57:20,447
ABSOLUTELY, IF YOU ARE PLANNING 

1569
00:57:20,447 --> 00:57:21,982
TO VALIDATE ANYTHING YOU FIND 

1570
00:57:21,982 --> 00:57:23,417
USING THIS DATASET YOU WILL NEED

1571
00:57:23,417 --> 00:57:26,053
TO DO A SEPARATE STUDY THAT 

1572
00:57:26,053 --> 00:57:28,488
LOOKS AT THIS IDEALLY IN A 

1573
00:57:28,488 --> 00:57:29,322
PROSPECTIVE FASHION.

1574
00:57:29,322 --> 00:57:30,457
>> THANK YOU.

1575
00:57:30,457 --> 00:57:32,826
THAT'S A VERY NICE WAY TO 

1576
00:57:32,826 --> 00:57:33,760
EXPLAIN IT.

1577
00:57:33,760 --> 00:57:35,328
>> THAT'S EXCELLENT.

1578
00:57:35,328 --> 00:57:37,798
THANK YOU, DR. LEE.

1579
00:57:37,798 --> 00:57:39,633
DR. EDWARDS, DID YOU HAVE A 

1580
00:57:39,633 --> 00:57:41,902
QUESTION FOR DR. LEE?

1581
00:57:41,902 --> 00:57:43,370
THANK YOU.

1582
00:57:43,370 --> 00:57:45,172
THE DIRECTOR OF OUR DIVISION OF 

1583
00:57:45,172 --> 00:57:46,206
EXTERNAL RESEARCH FOR OUR 

1584
00:57:46,206 --> 00:57:46,740
VIEWERS.

1585
00:57:46,740 --> 00:57:48,642
>> THANK YOU SO MUCH, DR.
LEE, 

1586
00:57:48,642 --> 00:57:50,510
FOR A WONDERFUL PRESENTATION.

1587
00:57:50,510 --> 00:57:52,612
OF COURSE, IT LOOKS SO SIMPLE 

1588
00:57:52,612 --> 00:57:53,780
FROM YOUR PRESENTATION, BUT I 

1589
00:57:53,780 --> 00:57:55,549
KNOW THAT IS NOT THE CASE.

1590
00:57:55,549 --> 00:57:59,052
SO MY QUESTION TO YOU IS WHAT 

1591
00:57:59,052 --> 00:58:01,188
HAS BEEN YOUR GREATEST CHALLENGE

1592
00:58:01,188 --> 00:58:03,523
TO PUT THIS WONDERFUL DATASET 

1593
00:58:03,523 --> 00:58:03,790
TOGETHER?

1594
00:58:03,790 --> 00:58:05,792
I KNOW YOU HAVE A WONDERFUL 

1595
00:58:05,792 --> 00:58:07,861
TEAM, BUT COULD YOU SHARE, I 

1596
00:58:07,861 --> 00:58:09,296
MEAN, FOR PEOPLE THAT MIGHT BE 

1597
00:58:09,296 --> 00:58:11,331
INTERESTED IN FOLLOWING THAT 

1598
00:58:11,331 --> 00:58:13,433
PATH, WHAT ARE SOME OF THE 

1599
00:58:13,433 --> 00:58:14,367
CHALLENGES THAT YOU'VE 

1600
00:58:14,367 --> 00:58:15,001
ENCOUNTERED THAT YOU ARE WILLING

1601
00:58:15,001 --> 00:58:17,737
TO SHARE?

1602
00:58:17,737 --> 00:58:18,271
>> YEAH.

1603
00:58:18,271 --> 00:58:20,941
SO ONE OF THE THINGS THAT WE ARE

1604
00:58:20,941 --> 00:58:22,943
TRYING VERY HARD TO DO IS TO DO 

1605
00:58:22,943 --> 00:58:24,778
THIS THE QUOTE-UNQUOTE RIGHT 

1606
00:58:24,778 --> 00:58:24,945
WAY.

1607
00:58:24,945 --> 00:58:30,383
AND WHAT I MEAN BY THAT IS, YOU 

1608
00:58:30,383 --> 00:58:32,752
KNOW, WE ARE NOT USING 

1609
00:58:32,752 --> 00:58:34,621
CONVENIENCE SAMPLING WHEN WE ARE

1610
00:58:34,621 --> 00:58:35,789
TRYING TO GET THIS 4,000 PEOPLE.

1611
00:58:35,789 --> 00:58:37,991
WE ARE NOT INVITING EVERYBODY 

1612
00:58:37,991 --> 00:58:39,092
WHO STEPS THROUGH THE DOOR AND 

1613
00:58:39,092 --> 00:58:41,061
ASKING THEM TO COME.

1614
00:58:41,061 --> 00:58:43,697
INSTEAD, WHAT WE'VE DONE IS 

1615
00:58:43,697 --> 00:58:45,699
WE'VE PROBED THESE THREE MEDICAL

1616
00:58:45,699 --> 00:58:47,667
HEALTH CARE SYSTEMS AND PULLED 

1617
00:58:47,667 --> 00:58:48,902
DOWN EVERY PATIENT THAT HAS BEEN

1618
00:58:48,902 --> 00:58:51,438
SEEN IN THE LAST FIVE YEARS.

1619
00:58:51,438 --> 00:58:55,442
AND WE LOOK AT THEIR DEMOGRAPHIC

1620
00:58:55,442 --> 00:58:58,445
PROFILE AND WE SEND THEM SORT OF

1621
00:58:58,445 --> 00:59:00,147
AN INVITATION LETTER THROUGH THE

1622
00:59:00,147 --> 00:59:03,083
MAIL OR E-MAIL TO INVITE THEM TO

1623
00:59:03,083 --> 00:59:04,684
BE PART OF THE STUDY.

1624
00:59:04,684 --> 00:59:05,986
THE REASON WE ARE DOING THIS IS 

1625
00:59:05,986 --> 00:59:07,988
BECAUSE WE WANT TO TRY TO GET 

1626
00:59:07,988 --> 00:59:11,892
RID OF THE HEALTHY VOLUNTEER 

1627
00:59:11,892 --> 00:59:12,959
BIAS AS MUCH AS POSSIBLE.

1628
00:59:12,959 --> 00:59:15,929
BUT WE ARE SETTING OURSELVES UP 

1629
00:59:15,929 --> 00:59:17,330
FOR AN UPHILL BATTLE FOR SURE.

1630
00:59:17,330 --> 00:59:20,233
THAT IS DEFINITELY THE HARDEST 

1631
00:59:20,233 --> 00:59:21,868
THING THAT WE, AS A PROJECT, 

1632
00:59:21,868 --> 00:59:23,203
HAVE BEEN DEALING WITH.

1633
00:59:23,203 --> 00:59:25,238
IS TRYING TO GET PEOPLE TO SAY 

1634
00:59:25,238 --> 00:59:25,939
YES.

1635
00:59:25,939 --> 00:59:27,440
NOW, YOU KNOW, WHAT HAS BEEN 

1636
00:59:27,440 --> 00:59:28,475
ENCOURAGING IS THAT WHEN WE LOOK

1637
00:59:28,475 --> 00:59:32,646
AT THE PERCENTAGES OF THE NUMBER

1638
00:59:32,646 --> 00:59:35,282
OF MALES, LETTERS WE HAVE SENT 

1639
00:59:35,282 --> 00:59:36,883
THROUGH THE MALE AND THE NUMBER 

1640
00:59:36,883 --> 00:59:39,352
OF PEOPLE RESPONDED YES AND 

1641
00:59:39,352 --> 00:59:40,787
COMPLETED THIS 10-DAY PROTOCOL 

1642
00:59:40,787 --> 00:59:42,789
OF WEARING THE SENSORS AND 

1643
00:59:42,789 --> 00:59:44,257
HAVING THAT ENVIRONMENTAL SENSOR

1644
00:59:44,257 --> 00:59:48,328
AND ALL THIS BLOOD DRAWN.

1645
00:59:48,328 --> 00:59:50,063
THEY ULTIMATELY ENDS UP BEING 

1646
00:59:50,063 --> 00:59:54,935
THAT THE PARTICIPANT RATE IS 

1647
00:59:54,935 --> 00:59:59,206
AROUND, I THINK, 6% OR 7% OR SO.

1648
00:59:59,206 --> 01:00:00,874
AND EVEN THOUGH THAT SOUNDS LOW,

1649
01:00:00,874 --> 01:00:02,242
GIVEN THE NUMBER OF PEOPLE WE 

1650
01:00:02,242 --> 01:00:04,144
HAVE IN OUR THREE COMBINED 

1651
01:00:04,144 --> 01:00:05,178
HEALTH CARE SYSTEMS, WE ARE 

1652
01:00:05,178 --> 01:00:10,951
ACCOUNT -- CONFIDENT WE WILL GET

1653
01:00:10,951 --> 01:00:12,852
TO THE 4,000 WE NEED.

1654
01:00:12,852 --> 01:00:14,354
TO ANSWER YOUR QUESTION THE 

1655
01:00:14,354 --> 01:00:16,056
HARDEST THING WE ARE FACING IS 

1656
01:00:16,056 --> 01:00:17,724
FINDING THE PEOPLE WHO ARE 

1657
01:00:17,724 --> 01:00:20,126
WILLING TO GO THROUGH THIS 

1658
01:00:20,126 --> 01:00:21,761
REALLY BURDENSOME PROTOCOL AND 

1659
01:00:21,761 --> 01:00:22,729
BE GRACIOUS ENOUGH TO SHARE 

1660
01:00:22,729 --> 01:00:24,698
THEIR DATA FOR THE SCIENTIFIC 

1661
01:00:24,698 --> 01:00:29,169
WORLD TO LEARN FROM.

1662
01:00:29,169 --> 01:00:29,803
>> THANK YOU.

1663
01:00:29,803 --> 01:00:31,972
>> WELL, WE DO HAVE A
COUPLE OF 

1664
01:00:31,972 --> 01:00:33,306
QUESTIONS FROM THE AUDIENCE.

1665
01:00:33,306 --> 01:00:37,477
WE HAVE THE FIRST ONE HERE, DR. 

1666
01:00:37,477 --> 01:00:37,744
LEE.

1667
01:00:37,744 --> 01:00:40,180
DO YOU EVER ENVISION USING 

1668
01:00:40,180 --> 01:00:43,183
DATASETS COMPRISED OF HEALTHY 

1669
01:00:43,183 --> 01:00:43,717
PEOPLE?

1670
01:00:43,717 --> 01:00:46,253
CAN AI BE USED TO STUDY HEALTH 

1671
01:00:46,253 --> 01:00:49,022
INSTEAD OF ONLY STUDYING 

1672
01:00:49,022 --> 01:00:49,256
DISEASE?

1673
01:00:49,256 --> 01:00:52,092
FOR EXAMPLE, OBVIOUSLY, YOU ARE 

1674
01:00:52,092 --> 01:00:54,928
STUDYING DIABETES OR MACULAR 

1675
01:00:54,928 --> 01:00:55,395
DEGENERATION.

1676
01:00:55,395 --> 01:00:58,131
THE PERSON IS ASKING CAN IT BE 

1677
01:00:58,131 --> 01:01:00,867
USED TO STUDY WHAT MOTIVATES 

1678
01:01:00,867 --> 01:01:04,170
PEOPLE TO MAKE HEALTHY LIFESTYLE

1679
01:01:04,170 --> 01:01:04,671
BEHAVIOR CHANGES?

1680
01:01:04,671 --> 01:01:05,572
>> YEAH.

1681
01:01:05,572 --> 01:01:08,908
SO ONE THING THAT I WANT TO 

1682
01:01:08,908 --> 01:01:11,244
STRESS IS THAT WE ARE IN OUR 

1683
01:01:11,244 --> 01:01:12,379
DATASET COLLECTING EVEN NUMBERS 

1684
01:01:12,379 --> 01:01:15,815
OF PEOPLE WHO ARE HEALTHY VERSUS

1685
01:01:15,815 --> 01:01:16,116
DISEASED.

1686
01:01:16,116 --> 01:01:20,153
SO ONE OF THE BIG PROBLEMS WITH 

1687
01:01:20,153 --> 01:01:22,088
DATASETS IN GENERAL IS THAT THEY

1688
01:01:22,088 --> 01:01:24,724
TEND TO BE EITHER ENRICHED FOR 

1689
01:01:24,724 --> 01:01:26,326
VERY HEAVILY ENRICHED FOR 

1690
01:01:26,326 --> 01:01:29,062
DISEASE OR VERY HEAVILY ENRICHED

1691
01:01:29,062 --> 01:01:29,996
FOR NORMAL.

1692
01:01:29,996 --> 01:01:32,899
I WILL KIND OF GIVE A COUPLE OF 

1693
01:01:32,899 --> 01:01:33,566
EXAMPLES FOR THAT.

1694
01:01:33,566 --> 01:01:35,835
A DATASET THAT YOU MAY BE 

1695
01:01:35,835 --> 01:01:38,305
FAMILIAR WITH IS THE UK BIOBANK.

1696
01:01:38,305 --> 01:01:40,407
IT IS A VERY LARGE DATASET 

1697
01:01:40,407 --> 01:01:42,208
COLLECTED IN THE UNITED KINGDOM.

1698
01:01:42,208 --> 01:01:44,110
IT IS VERY SUCCESSFUL AND THEY 

1699
01:01:44,110 --> 01:01:45,278
HAVE COLLECTED ABOUT HALF A 

1700
01:01:45,278 --> 01:01:48,682
MILLION PEOPLE IN THAT DATASET.

1701
01:01:48,682 --> 01:01:49,849
NOW THE PROBLEM IS THAT THEY ARE

1702
01:01:49,849 --> 01:01:52,886
ALL VERY, VERY HEALTHY.

1703
01:01:52,886 --> 01:01:54,287
AND SO, IN FACT, LIKE COMMON 

1704
01:01:54,287 --> 01:01:57,123
DISEASES ARE ACTUALLY FAIRLY 

1705
01:01:57,123 --> 01:01:59,793
UNCOMMON IN THE UK BIOBANK.

1706
01:01:59,793 --> 01:02:01,227
SO THAT MAKES IT VERY 

1707
01:02:01,227 --> 01:02:03,196
CHALLENGING TO ACTUALLY 

1708
01:02:03,196 --> 01:02:05,665
UNDERSTAND HOW, IF SOMEBODY 

1709
01:02:05,665 --> 01:02:07,667
STARTS OUT BEING EXTRAORDINARILY

1710
01:02:07,667 --> 01:02:09,836
HEALTHY, THERE IS NOT ROOM FOR 

1711
01:02:09,836 --> 01:02:11,671
THEM TO BECOME HEALTHIER.

1712
01:02:11,671 --> 01:02:14,708
THERE IS SORT OF A CEILING 

1713
01:02:14,708 --> 01:02:14,941
EFFECT.

1714
01:02:14,941 --> 01:02:16,876
YOU WANT A SPREAD, A SPECTRUM OF

1715
01:02:16,876 --> 01:02:20,080
PEOPLE, ALONG, IDEALLY ALONG 

1716
01:02:20,080 --> 01:02:23,750
THAT DISEASE AX IS.

1717
01:02:23,750 --> 01:02:26,119
THAT IS WHY IN OUR DATASET WE 

1718
01:02:26,119 --> 01:02:27,954
HAVE 1,000 PEOPLE THAT HAVE NO 

1719
01:02:27,954 --> 01:02:28,555
DIABETES.

1720
01:02:28,555 --> 01:02:30,256
THEY ARE NOT EVEN PREDIABETES.

1721
01:02:30,256 --> 01:02:31,758
THERE IS ABSOLUTELY, THEY 

1722
01:02:31,758 --> 01:02:34,294
BELIEVE THEY ARE LIVING A 

1723
01:02:34,294 --> 01:02:35,528
HEALTHY LIFESTYLE AND THEY DO 

1724
01:02:35,528 --> 01:02:37,063
NOT HAVE ANY DIABETES.

1725
01:02:37,063 --> 01:02:41,000
AND THEN THERE'S 1,000 OF THE 

1726
01:02:41,000 --> 01:02:50,610
PEOPLE, INDEED, PREDIABETIC AND 

1727
01:02:50,610 --> 01:02:55,615
ARE NOT STARTED ON MEDICATIONS 

1728
01:02:55,615 --> 01:02:56,349
YET.

1729
01:02:56,349 --> 01:02:58,017
A THOUSAND OF THE FOLKS ARE 

1730
01:02:58,017 --> 01:03:01,087
PEOPLE ON DRUGS TO TRY AND 

1731
01:03:01,087 --> 01:03:03,790
CONTROL THEIR DIABETES AND THE 

1732
01:03:03,790 --> 01:03:06,559
LAST GROUP IS PEOPLE WHO REQUIRE

1733
01:03:06,559 --> 01:03:09,662
INSULIN TO SORT OF STAY ALIVE 

1734
01:03:09,662 --> 01:03:10,697
WITH THEIR TYPE 2 DIABETES.

1735
01:03:10,697 --> 01:03:14,267
SO WE ARE TRYING TO SAMPLE EVEN 

1736
01:03:14,267 --> 01:03:16,202
NUMBERS OF THOSE FOUR DIFFERENT 

1737
01:03:16,202 --> 01:03:18,037
GROUPS AND TO CREATE THIS SPREAD

1738
01:03:18,037 --> 01:03:22,642
OF PEOPLE SO THAT YOU CAN STUDY 

1739
01:03:22,642 --> 01:03:24,844
THINGS LIKE WHAT KINDS OF 

1740
01:03:24,844 --> 01:03:26,379
LIFESTYLE CHANGES MIGHT MOTIVATE

1741
01:03:26,379 --> 01:03:27,881
PEOPLE TO BECOME HEALTHIER.

1742
01:03:27,881 --> 01:03:30,316
>> THANK YOU SO MUCH.

1743
01:03:30,316 --> 01:03:32,118
AND THE SECOND QUESTION WE 

1744
01:03:32,118 --> 01:03:36,189
RECEIVED FROM AN AUDIENCE 

1745
01:03:36,189 --> 01:03:37,657
MEMBER, THE AUDIENCE MEMBER SAYS

1746
01:03:37,657 --> 01:03:40,126
I THOUGHT THE MACHINE LEARNING 

1747
01:03:40,126 --> 01:03:42,462
MODELS ARE LIKE BLACK BOXES AND 

1748
01:03:42,462 --> 01:03:45,098
CANNOT REVEAL HOW THE DATA IS 

1749
01:03:45,098 --> 01:03:45,632
ANALYZED.

1750
01:03:45,632 --> 01:03:47,233
CAN YOU PLEASE NAME SOME MODELS 

1751
01:03:47,233 --> 01:03:49,235
YOU ARE ABLE TO SHARE REGARDING 

1752
01:03:49,235 --> 01:03:54,808
HOW THE DATA WAS ANALYZED TO 

1753
01:03:54,808 --> 01:03:55,675
FINDINGS.

1754
01:03:55,675 --> 01:03:58,211
ECG BIOMARKERS AND THE PERSON 

1755
01:03:58,211 --> 01:03:59,179
SAYS THANK YOU FOR A WONDERFUL 

1756
01:03:59,179 --> 01:03:59,379
TALK.

1757
01:03:59,379 --> 01:04:01,080
>> YEAH.

1758
01:04:01,080 --> 01:04:03,817
I'M REGRETTING CUTTING -- THERE 

1759
01:04:03,817 --> 01:04:06,186
IS AN EARLIER VERSION OF THIS 

1760
01:04:06,186 --> 01:04:08,421
SLIDE FIVE HOURS AGO WHERE I HAD

1761
01:04:08,421 --> 01:04:11,357
FIVE SLIDES SHOWING THIS 

1762
01:04:11,357 --> 01:04:12,125
VISUALIZATION TECHNIQUE.

1763
01:04:12,125 --> 01:04:15,228
I CUT THAT OUT.

1764
01:04:15,228 --> 01:04:16,362
I ASSUMED THIS AUDIENCE MAY NOT 

1765
01:04:16,362 --> 01:04:19,199
BE INTERESTED IN IMAGE ANALYSIS 

1766
01:04:19,199 --> 01:04:19,499
TECHNIQUES.

1767
01:04:19,499 --> 01:04:21,568
I'M REGRETTING THAT IN THIS 

1768
01:04:21,568 --> 01:04:22,268
MOMENT.

1769
01:04:22,268 --> 01:04:24,537
IT WOULD HAVE BEEN REALLY 

1770
01:04:24,537 --> 01:04:29,776
HELPFUL TO ELUCIDATE HOW THIS 

1771
01:04:29,776 --> 01:04:30,510
WORKS.

1772
01:04:30,510 --> 01:04:31,644
THE PERSON ASKING THIS QUESTION 

1773
01:04:31,644 --> 01:04:35,081
IS ABSOLUTELY CORRECT.

1774
01:04:35,081 --> 01:04:37,083
DEEP LEARNING MODELS AND COMPLEX

1775
01:04:37,083 --> 01:04:38,418
MACHINE LEARNING MODELS HAVE A 

1776
01:04:38,418 --> 01:04:40,887
PROBLEM WITH INTERPRETABILITY OR

1777
01:04:40,887 --> 01:04:41,254
EXPLAINABILITY.

1778
01:04:41,254 --> 01:04:43,223
YOU CAN'T FULLY UNDERSTAND HOW A

1779
01:04:43,223 --> 01:04:45,992
MODEL IS ABLE TO MAKE ITS 

1780
01:04:45,992 --> 01:04:47,494
DECISION OR COME TO ITS FINAL 

1781
01:04:47,494 --> 01:04:51,898
CONCLUSIONS AND YOU CAN'T REALLY

1782
01:04:51,898 --> 01:04:54,033
TAKE IT APART AND FOLLOW EVERY 

1783
01:04:54,033 --> 01:04:56,102
PART OF THAT LOGIC PATTERN 

1784
01:04:56,102 --> 01:04:57,871
BECAUSE IT IS SO COMPLEX.

1785
01:04:57,871 --> 01:05:00,673
IT IS THE REASON WHY THEY ARE SO

1786
01:05:00,673 --> 01:05:04,344
POWERFUL AND ABLE TO DO COMPLEX 

1787
01:05:04,344 --> 01:05:05,645
THINGS COME FROM THAT HIGH -- 

1788
01:05:05,645 --> 01:05:06,779
SPACE THEY WORK IN.

1789
01:05:06,779 --> 01:05:08,548
THERE IS AN INHERENT PROBLEM 

1790
01:05:08,548 --> 01:05:10,650
WITH UNDERSTANDING 100% OF HOW A

1791
01:05:10,650 --> 01:05:15,455
MODEL IS ABLE TO DO ANYTHING.

1792
01:05:15,455 --> 01:05:19,826
NOW THERE ARE A CLASS OF 

1793
01:05:19,826 --> 01:05:21,060
POSTTALK METHODS THAT CAN BE 

1794
01:05:21,060 --> 01:05:22,862
USED TO PROBE HOW THE MODEL IS 

1795
01:05:22,862 --> 01:05:24,464
MAKING ITS DECISION.

1796
01:05:24,464 --> 01:05:28,368
IN THE SPACE OF IMAGE ANALYSIS, 

1797
01:05:28,368 --> 01:05:30,403
ONE OF THE EXAMPLES WE USED IN 

1798
01:05:30,403 --> 01:05:32,505
THE FIRST PART I GAVE IS 

1799
01:05:32,505 --> 01:05:34,440
COVERING UP A SMALL PART OF THAT

1800
01:05:34,440 --> 01:05:36,609
IMAGE AND PASSING IT THROUGH THE

1801
01:05:36,609 --> 01:05:38,711
MODEL AND SEE HOW MUCH COVERING 

1802
01:05:38,711 --> 01:05:40,446
UP THAT PART OF THE IMAGE 

1803
01:05:40,446 --> 01:05:42,048
INFLUENCED THE MODEL'S ABILITY 

1804
01:05:42,048 --> 01:05:43,750
TO MAKE ITS DETERMINATION.

1805
01:05:43,750 --> 01:05:46,753
IF YOU COVER UP A CRITICAL PART 

1806
01:05:46,753 --> 01:05:48,755
OF THAT IMAGE THE MODEL WILL 

1807
01:05:48,755 --> 01:05:51,024
FAIL IN BEING ABLE TO TELL IF IT

1808
01:05:51,024 --> 01:05:53,226
IS NORMAL VERSUS MACULAR 

1809
01:05:53,226 --> 01:05:53,560
DEGENERATION.

1810
01:05:53,560 --> 01:05:54,827
AND THAT IS HOW YOU CAN TELL 

1811
01:05:54,827 --> 01:05:56,896
THAT PART OF THE IMAGE IS 

1812
01:05:56,896 --> 01:05:58,264
REALLY, REALLY IMPORTANT FOR THE

1813
01:05:58,264 --> 01:05:59,399
MODEL TO DO THAT.

1814
01:05:59,399 --> 01:06:00,567
THE OTHER EXAMPLE THAT I CAN 

1815
01:06:00,567 --> 01:06:03,102
GIVE YOU IS THAT A BEAUTIFUL 

1816
01:06:03,102 --> 01:06:05,071
VIDEO THAT I STOLE FROM FACEBOOK

1817
01:06:05,071 --> 01:06:08,541
THAT HAD THE RUNNING DOG AND THE

1818
01:06:08,541 --> 01:06:08,908
HORSE.

1819
01:06:08,908 --> 01:06:10,577
AND THE MODEL WAS SORT OF 

1820
01:06:10,577 --> 01:06:13,980
SHOWING WHERE IT WAS LOOKING.

1821
01:06:13,980 --> 01:06:17,283
SO THAT FAMILY OF DEEP LEARNING 

1822
01:06:17,283 --> 01:06:20,086
MODELS ARE USED WITH SOMETHING 

1823
01:06:20,086 --> 01:06:22,088
CALLED ATTENTION AND THEY HAVE 

1824
01:06:22,088 --> 01:06:23,556
THESE DIFFERENT ATTENTION HEADS 

1825
01:06:23,556 --> 01:06:25,191
THAT YOU CAN PROBE DIRECTLY TO 

1826
01:06:25,191 --> 01:06:26,659
UNDERSTAND WHAT PART OF THE 

1827
01:06:26,659 --> 01:06:29,028
IMAGE IS BEING USED.

1828
01:06:29,028 --> 01:06:31,397
AND THOSE VIDEOS ARE ACTUALLY 

1829
01:06:31,397 --> 01:06:34,267
BEING COLORED DIRECTLY FROM THE 

1830
01:06:34,267 --> 01:06:36,436
ATTENTION HEADS THAT ARE PART OF

1831
01:06:36,436 --> 01:06:38,938
THAT MODEL THAT WAS TRAINED IN 

1832
01:06:38,938 --> 01:06:41,874
THE DINO FASHION.

1833
01:06:41,874 --> 01:06:44,844
SO, YES, IT IS ABSOLUTELY TRUE 

1834
01:06:44,844 --> 01:06:46,312
THAT WE CAN'T UNDERSTAND HOW 

1835
01:06:46,312 --> 01:06:48,147
THESE MODELS WORK AND WE CAN'T 

1836
01:06:48,147 --> 01:06:49,749
UNDERSTAND HOW IT IS MAKING A 

1837
01:06:49,749 --> 01:06:51,017
PARTICULAR DECISION, BUT THERE 

1838
01:06:51,017 --> 01:06:54,454
IS SOME HOPE OF SOME LEVEL OF 

1839
01:06:54,454 --> 01:06:56,923
EXPLAINABILITY WHERE WE CAN GET 

1840
01:06:56,923 --> 01:06:58,758
HINTS OUT FOR HOW THE MODEL IS 

1841
01:06:58,758 --> 01:06:59,759
ABLE TO DO ITS TASK.

1842
01:06:59,759 --> 01:07:02,095
IF YOU SCALE THAT UP TO A LARGE 

1843
01:07:02,095 --> 01:07:05,898
NUMBER OF EXAMPLES THAT ARE 

1844
01:07:05,898 --> 01:07:07,500
CONSISTENT AND BEHAVES IN A 

1845
01:07:07,500 --> 01:07:08,868
CONSISTENT FASHION, YOU CAN 

1846
01:07:08,868 --> 01:07:11,070
START TO DISCOVER NEW BIOMARKERS

1847
01:07:11,070 --> 01:07:13,373
FROM THESE MODELS.

1848
01:07:13,373 --> 01:07:15,308
>> THANK YOU SO MUCH.

1849
01:07:15,308 --> 01:07:17,510
DAVID, DID YOU HAVE ANY 

1850
01:07:17,510 --> 01:07:20,580
FOLLOW-UP QUESTION FROM EARLIER,

1851
01:07:20,580 --> 01:07:23,316
ANYTHING THAT YOU WANTED TO 

1852
01:07:23,316 --> 01:07:25,284
CIRCLE BACK TO BEFORE WE START 

1853
01:07:25,284 --> 01:07:26,419
TO CONCLUDE?

1854
01:07:26,419 --> 01:07:26,586
IS 

1855
01:07:26,586 --> 01:07:26,986
>> YEAH.

1856
01:07:26,986 --> 01:07:29,489
YOU HAVE TOUCHED A LOT OF ISSUES

1857
01:07:29,489 --> 01:07:31,391
I HAVE QUESTIONS ABOUT.

1858
01:07:31,391 --> 01:07:34,594
ONE ISSUE IS YOU SORT OF 

1859
01:07:34,594 --> 01:07:45,071
MENTIONED AND HE HELENE MENTION.

1860
01:07:45,071 --> 01:07:46,906
I'M WONDERING IF YOU CAN GIVE US

1861
01:07:46,906 --> 01:07:48,474
A SENSE OF HOW YOU SEE THE 

1862
01:07:48,474 --> 01:07:50,643
FUTURE IN TERMS OF -- I ALWAYS 

1863
01:07:50,643 --> 01:07:52,345
FEEL LIKE WHAT WE ARE DOING WHEN

1864
01:07:52,345 --> 01:07:54,647
WE TREAT DOES IT IS REACTIVE 

1865
01:07:54,647 --> 01:07:55,982
RATHER THAN PROACTIVE.

1866
01:07:55,982 --> 01:07:58,117
I THINK THE CONTINUUM YOU SHOWED

1867
01:07:58,117 --> 01:08:00,953
FROM HEALTH TO DISEASE AND THE 

1868
01:08:00,953 --> 01:08:01,821
WAY MACHINE LEARNING CAN GET 

1869
01:08:01,821 --> 01:08:05,491
DEEPER IN FINDING PATTERNS THAT 

1870
01:08:05,491 --> 01:08:07,660
MAY BE LESS -- HUMANS ARE LESS 

1871
01:08:07,660 --> 01:08:09,095
ABLE TO PIECE TOGETHER.

1872
01:08:09,095 --> 01:08:12,699
ARE THERE WAYS YOU SEE HOW THIS 

1873
01:08:12,699 --> 01:08:13,833
REALLY COULD REVOLUTIONIZE 

1874
01:08:13,833 --> 01:08:14,267
HEALTH CARE?

1875
01:08:14,267 --> 01:08:16,502
I MEAN, CAN YOU TALK ABOUT MAYBE

1876
01:08:16,502 --> 01:08:19,539
HOW WE SHIFT THE MODEL, A REAL 

1877
01:08:19,539 --> 01:08:20,440
PARADIGM SHIFT IN HEALTH CARE, 

1878
01:08:20,440 --> 01:08:22,175
HOW THIS DATA MIGHT ADDRESS SOME

1879
01:08:22,175 --> 01:08:24,644
OF THE REAL NEEDS IN HEALTH 

1880
01:08:24,644 --> 01:08:25,111
CARE?

1881
01:08:25,111 --> 01:08:27,113
COULD YOU JUST SPEAK TO THAT?

1882
01:08:27,113 --> 01:08:27,914
>> YEAH.

1883
01:08:27,914 --> 01:08:29,615
SO I THINK THESE FAMILY OF 

1884
01:08:29,615 --> 01:08:32,685
ANALYSIS TOOLS CAN REALLY BE 

1885
01:08:32,685 --> 01:08:36,956
USED TO SORT OF REENGINEER THE 

1886
01:08:36,956 --> 01:08:38,858
WAY THAT PEOPLE HAVE DONE 

1887
01:08:38,858 --> 01:08:39,992
SCIENTIFIC DISCOVERY.

1888
01:08:39,992 --> 01:08:44,363
WHAT I MEAN BY THAT IS USUALLY 

1889
01:08:44,363 --> 01:08:46,933
THERE IS SOME SORT OF 

1890
01:08:46,933 --> 01:08:53,072
OBSERVATION AND PEOPLE 

1891
01:08:53,072 --> 01:08:56,776
DEVELOPMENT HIYPOTHESIS AND PROE

1892
01:08:56,776 --> 01:09:01,080
THE HYPOTHESIS AND THAT IS THE 

1893
01:09:01,080 --> 01:09:01,714
DISCOVERY.

1894
01:09:01,714 --> 01:09:03,616
SOMETIMES THOSE OBSERVATIONS 

1895
01:09:03,616 --> 01:09:04,250
REQUIRE HUMAN BEINGS BEING 

1896
01:09:04,250 --> 01:09:07,286
LOCKED UP IN A ROOM WITH EXPERTS

1897
01:09:07,286 --> 01:09:09,322
ALL COMING TO A CONSENSUS ABOUT 

1898
01:09:09,322 --> 01:09:11,157
SOMETHING OR BREAKING DOWN THE 

1899
01:09:11,157 --> 01:09:12,592
DISEASE INTO DIFFERENT STAGES.

1900
01:09:12,592 --> 01:09:15,528
THAT IS ANOTHER EXAMPLE OF THAT.

1901
01:09:15,528 --> 01:09:19,332
AND I KIND OF VIEW THAT THERE 

1902
01:09:19,332 --> 01:09:20,900
MIGHT BE A HOPE WHERE THESE 

1903
01:09:20,900 --> 01:09:22,935
KINDS OF DATASETS AND THESE 

1904
01:09:22,935 --> 01:09:26,205
TYPES OF METHODS COULD BE USED 

1905
01:09:26,205 --> 01:09:29,208
TO GIVE US A NEW WAY TO DISCOVER

1906
01:09:29,208 --> 01:09:32,845
ASSOCIATIONS THAT ARE NOT BEING 

1907
01:09:32,845 --> 01:09:34,280
SO BIASED BY HUMAN 

1908
01:09:34,280 --> 01:09:36,015
UNDERSTANDING, RIGHT?

1909
01:09:36,015 --> 01:09:38,785
IF YOU LOOK IN PUB MED THERE ARE

1910
01:09:38,785 --> 01:09:40,720
CERTAIN PATHWAYS PROBED OVER AND

1911
01:09:40,720 --> 01:09:41,687
OVER AND OVER AGAIN.

1912
01:09:41,687 --> 01:09:44,023
THERE IS A WHOLE BODY OF 

1913
01:09:44,023 --> 01:09:44,323
LITERATURE.

1914
01:09:44,323 --> 01:09:48,060
IN ACTUALITY THERE IS MANY, MANY

1915
01:09:48,060 --> 01:09:50,797
MORE BIOCHEMICAL PATHWAYS THAT 

1916
01:09:50,797 --> 01:09:52,498
EXIST IN THE BODY AND THE 

1917
01:09:52,498 --> 01:09:54,100
INTERACTION WITH THE ENVIRONMENT

1918
01:09:54,100 --> 01:09:56,302
IS MORE COMPLEX THAN HOW DOES 

1919
01:09:56,302 --> 01:09:58,404
THIS ONE PROTEIN INTERACT WITH 

1920
01:09:58,404 --> 01:10:01,874
THIS OTHER PROTEIN IN SOME SORT 

1921
01:10:01,874 --> 01:10:04,143
OF BIOCHEMICAL PATHWAY.

1922
01:10:04,143 --> 01:10:06,279
THESE METHODS GIVE US A WAY TO 

1923
01:10:06,279 --> 01:10:08,748
APPROACH DATA IN THE UNBIASED 

1924
01:10:08,748 --> 01:10:10,850
FASHION, WITHOUT THE LENS OF THE

1925
01:10:10,850 --> 01:10:13,753
HUMANS DOING EXPERIMENTS AND 

1926
01:10:13,753 --> 01:10:14,954
TRYING TO LOOK AT ASSOCIATIONS 

1927
01:10:14,954 --> 01:10:16,022
AND PATH WAYS.

1928
01:10:16,022 --> 01:10:20,193
IT GIVES US A DE NOVO WAY TO 

1929
01:10:20,193 --> 01:10:22,695
LOOK AT THIS DATASET WITHOUT 

1930
01:10:22,695 --> 01:10:24,230
THOSE BIASES BUILT IN.

1931
01:10:24,230 --> 01:10:26,299
THAT CAN BE VERY POWERFUL AND 

1932
01:10:26,299 --> 01:10:28,234
LEAD TO NEW DISCOVERIES THAT WE 

1933
01:10:28,234 --> 01:10:31,037
WERE, AS A SPECIES, JUST NOT 

1934
01:10:31,037 --> 01:10:32,038
THINKING ABOUT IN THAT 

1935
01:10:32,038 --> 01:10:35,341
PARTICULAR MOMENT.

1936
01:10:35,341 --> 01:10:37,910
>> GREAT.

1937
01:10:37,910 --> 01:10:39,512
THANKS SO MUCH.

1938
01:10:39,512 --> 01:10:41,480
THANK YOU SO MUCH, DR. LEE.

1939
01:10:41,480 --> 01:10:45,084
I THINK AT THIS POINT IN TIME WE

1940
01:10:45,084 --> 01:10:46,919
ARE GOING TO BRING THESE TO 

1941
01:10:46,919 --> 01:10:47,220
CONCLUSION.

1942
01:10:47,220 --> 01:10:50,289
I WANT TO THANK THE FOLKS WHO 

1943
01:10:50,289 --> 01:10:55,194
SENT IN SOME QUESTIONS, AND 

1944
01:10:55,194 --> 01:11:01,834
DAVID, HELENE AND EMMALINE.

1945
01:11:01,834 --> 01:11:05,071
DR. LEE, I APPRECIATE THE TALK 

1946
01:11:05,071 --> 01:11:07,673
AND BREAKING DOWN THE WAYS 

1947
01:11:07,673 --> 01:11:09,475
PEOPLE CAN START TO UNDERSTAND 

1948
01:11:09,475 --> 01:11:11,711
AI AND MACHINE LEARNING AND DEEP

1949
01:11:11,711 --> 01:11:13,112
LEARNING AND I REALLY APPRECIATE

1950
01:11:13,112 --> 01:11:15,681
YOU TAKING THE TIME TO DO THAT.

1951
01:11:15,681 --> 01:11:17,283
I THINK OUR AUDIENCE WILL 

1952
01:11:17,283 --> 01:11:19,118
DEFINITELY APPRECIATE IT AND 

1953
01:11:19,118 --> 01:11:21,487
THOSE WHO COME BACK AND WATCH AT

1954
01:11:21,487 --> 01:11:22,655
A LATER DATE.

1955
01:11:22,655 --> 01:11:25,024
WITH THAT SAID, I WANT TO REMIND

1956
01:11:25,024 --> 01:11:27,960
EVERYBODY THIS WAS RECORDED.

1957
01:11:27,960 --> 01:11:30,663
IT WILL BE IN THE NIH VIDEOCAST 

1958
01:11:30,663 --> 01:11:33,165
PAST EVENTS ARCHIVES WHERE YOU 

1959
01:11:33,165 --> 01:11:34,200
CAN JOIN AGAIN.

1960
01:11:34,200 --> 01:11:36,202
I WANT TO THANK EVERYONE FOR 

1961
01:11:36,202 --> 01:11:38,070
JOINING ON VIDEOCAST TO SEE THIS

1962
01:11:38,070 --> 01:11:39,238
GREAT PRESENTATION.

1963
01:11:39,238 --> 01:11:40,072
AND, OF COURSE, THANK YOU SO 

1964
01:11:40,072 --> 01:11:41,307
MUCH DR. LEE FOR YOUR TIME 

1965
01:11:41,307 --> 01:11:41,574
TODAY.

1966
01:11:41,574 --> 01:11:44,310
WITH THAT, I THINK WE CAN BRING 

1967
01:11:44,310 --> 01:11:46,345
OUR DAY TO A CONCLUSION.

1968
01:11:46,345 --> 01:11:49,548
DAVID AND HELENE, ANY FINAL 

1969
01:11:49,548 --> 01:11:51,384
THOUGHTS AS WE WRAP UP?

1970
01:11:51,384 --> 01:11:53,653
>> DAVID, ANY FINAL
THOUGHTS?

1971
01:11:53,653 --> 01:11:57,290
>> I THINK YOU SAID IT
ALL, 

1972
01:11:57,290 --> 01:11:57,556
CATHERINE.

1973
01:11:57,556 --> 01:11:58,691
THANK YOU, DR. LEE.

1974
01:11:58,691 --> 01:12:00,092
AGAIN, I THINK YOU ARE SHOWING 

1975
01:12:00,092 --> 01:12:01,527
US WHERE THE FUTURE IS AND I 

1976
01:12:01,527 --> 01:12:02,762
APPRECIATE WHAT YOU ARE DOING 

1977
01:12:02,762 --> 01:12:04,130
FOR THE PUBLIC HEALTH.

1978
01:12:04,130 --> 01:12:04,897
SO THANK YOU.

1979
01:12:04,897 --> 01:12:07,833
>> I THANK ALL OF YOU FOR
GIVING

1980
01:12:07,833 --> 01:12:09,035
ME THE OPPORTUNITY TO SPEAK 

1981
01:12:09,035 --> 01:12:09,402
TODAY.

1982
01:12:09,402 --> 01:12:12,271
IT IS A GREAT HONOR.

1983
01:12:12,271 --> 01:12:13,439
ANY FINAL THOUGHTS?

1984
01:12:13,439 --> 01:12:15,474
>> JUST TO SAY THANK YOU
FOR A 

1985
01:12:15,474 --> 01:12:17,109
WONDERFUL TALK AND I LOOK 

1986
01:12:17,109 --> 01:12:19,345
FORWARD TO TALKING TO YOU A 

1987
01:12:19,345 --> 01:12:19,712
LITTLE LATER.

1988
01:12:19,712 --> 01:12:21,647
>> SOUNDS GOOD.

1989
01:12:21,647 --> 01:12:26,519
>> I WANT TO CONGRATULATE
YOU ON

1990
01:12:26,519 --> 01:12:29,722
WHAT REALLY IS I'M SURE A VERY 

1991
01:12:29,722 --> 01:12:31,490
CHALLENGING PROJECT YOU HAVE 

1992
01:12:31,490 --> 01:12:33,726
TAKEN ON FEARLESSLY.

1993
01:12:33,726 --> 01:12:35,528
WE WISH YOU THE BEST OF LUCK 

1994
01:12:35,528 --> 01:12:36,762
WITH THE NEXT PHASE OF IT.

1995
01:12:36,762 --> 01:12:37,897
>> THANK YOU SO MUCH.

1996
01:12:37,897 --> 01:12:39,498
>> WELL, THANK YOU.

1997
01:12:39,498 --> 01:12:41,500
AND THANK YOU TO ALL OF OUR 

1998
01:12:41,500 --> 01:12:43,536
VIEWERS AND WITH THAT I'M NOW 

1999
01:12:43,536 --> 01:12:47,406
GOING TO CONCLUDE TODAY'S 

2000
01:12:47,406 --> 01:12:47,740
PRESENTATION.

2001
01:12:47,740 --> 01:12:48,541
BYE-BYE, EVERYBODY.

2002
01:12:48,541 --> 01:12:58,951
HAVE A GREAT AFTERNOON.