ministry of internal affairs and communications4.decisions referred to in paragraph 2 shall not be...
TRANSCRIPT
���������������
�
���� �����
1
���
6�3�
n�5<� P��Q• ~2013.3, PhD@;�., :�• 2013.4~2016.3, .0�@IBM!��/.• 2016.4~2017.8, .0�@&�"ERATO, NII• 2017.9~, ��@;�., :�
n.0• *�9�
� DLHBCMJGM>$8�4 PECML’11Q� ,���>�PAISTATS’15,17Q
• %#�4JGM>7�� ANENIM >1+� PAISTATS’18Q� JGM��=?@KOF>2��� PAAAI’17,18Q� (��4JGM>'-)>�� PongoingQ
2
/2;HE – 1n�$�:@ *AI+ = *���QMT+I��;49 I%AF)• ���QMTZKVNRWL?OUJSP:(�>��?��MWLIB;>,F��?��Xe.g. ��!"� Y>8-9�&�3G7B?)
• ���QMT?�� �;�?��5G6G100�0D(�'� 1�>=FC.>�&�3G7�����'�)
n *��AI+ =<@�#�)
3
/3;FD – 2n�#�:? +AI0MPKHNKHI8;�E��,G�"=�G$AE*
n +AI, B +AI> �,B)-2@:B!(��>7A>LSQR��*&�&�0��*• AI0MPKHNKHI:B�C<.��=@:) �G�AE �?<.*
• + �:1<.MPKHNKHI<AI?�9JO, ;�5E��?<.*%�=�G�'6E>?�@42<.*
4
AI�����������AI�� ���������
WQY
n('9;AI_5!I] #<T,-U����S?]D9+�*SkmcT��T4$VU/DLO@R@:
n,-D�T�TgmdQdlO@]�1�D?]:• abfjaT,-0T;F`R5!?LK\�H[R@n<Q@A�5S�NE5!#T,-D�@:
n(�C\T;gmdT)�<D��:• %&QGO��_9��*R�8S2QG7XFQP9�3*S4"�1CZG^R@:
• ��TWW�B7`PZ96ZheimSR\R@:
n WJU9=5!#T� D�.LK��*R�>_�M�LO9�T�S)�GK@:
5
������
n �AI�����������
n �AI���������
n �AI������������
6
��)AI*��#��n� (%'
• �!�&)��-��
n��(%'• ��-��)����)��
7
����#��&$(! ��+)�� ���)��
�����
��
��
�����
…
��,�")�!
��2AI3��*��n� 1,0
• !'�/2��9��
n��1,0• ��9��2����2��
8
�����
��
XX�����
XX�����
���
����*��/+7& $%42� ��2��� #"��2��
,816�-.�(7)5
�=]��(�2 \AIg�I\^7!K�3n-']6 ^E��K�QFb]D
• AIY��X6 KBH:VUcr� AI]� #�KfJc[HYCTbTb��]%5bXL[HD� #��!CYAIg*4Qd[cCTbTbAI�]noklK[HD
n<1ipmj��/^EA�_]7!].�FKGdD• 8g>NUa\^�$[�"K�?DAI]���;D• EG[U\^19XL`RhFE[SrF
� +,g7!PWMe[Hipmj�&/^�@Oe[HD
nT]�q�)C�0C[Z[Z
9
]AIn?"^qn3�0l:A
n�-0m]QUn?"^aG:<dvj_u\
• !#� QU�),��%�6�1LKOMNP�� E"�n��
• QU�������z��p�����,7o[QU�����oQU����n��n&>�8��p 5$n?"�8�m.�fu\
� x}{�����y�?"C��n��• AI�������z��p�����,7o[*D70�,7�pI��,7w�r���~���m�ex}{�����yw$gft`�su\
• SY� �9����B;��TRVW�� ���m�ic��(�mh_j[���n'�w�BeF�l�w�>fuC�w������7m@f\
• YX� ?"�8QU�ZQULKRQWVQ���|~��� ]�Ia+=eJkbuQU^n24H/\
10
�� EU��� ������GDPR�n GDPR-221. The data subject shall have the right not to be subject to a decision based
solely on automated processing, including profiling, which produces legal effects concerning him or her or similarly significantly affects him or her.
2. Paragraph 1 shall not apply if the decision: is necessary for entering into, or performance of, a contract between the data subject and a data controller; is authorised by Union or Member State law to which the controller is subject and which also lays down suitable measures to safeguard the data subject’s rights and freedoms and legitimate interests; or is based on the data subject’s explicit consent.
3. In the cases referred to in points (a) and (c) of paragraph 2, the data controller shall implement suitable measures to safeguard the data subject’s rights and freedoms and legitimate interests, at least the right to obtain human intervention on the part of the controller, to express his or her point of view and to contest the decision.
4. Decisions referred to in paragraph 2 shall not be based on special categories of personal data referred to in Article 9(2)1), unless point (a) or (g) of Article 9(2) applies and suitable measures to safeguard the data subject’s rights and freedoms and legitimate interests are in place.
12
������
n �AI�����������
n �AI���������
n �AI������������
13
���3��
n 2016��#%���" 3�$��/'AI3��(2"-5��*�• ICML, NIPS103���3;:<��/4'AI3��(2"-5>?89=:<*!�,6.)5&
n 2016���/%GDPR2�-5��3 ��7�+��*��,6.)5&
14
https://arxiv.org/abs/1606.08813
�����
n �AI����� �������
15
Peeking inside the black-box: A survey on Explainable Artificial Intelligence (XAI)
https://ieeexplore.ieee.org/document/8466590/
7� ������SCOPUS,
IEEExplore, ACM Digital Library, Google
Scholar, Citeseer Library, ScienceDirect,
arXiv����"%��&�������� “intelligible”,
“interpretable”, “transparency”, “black box”,
“understandable”, “comprehensible”,
“explainable”���AI�& ��“Artificial Intelligence”, “Intelligent
system”, “Machine learning”, “deep learning”,
“classifier” , “decision tree”���# ���'$ !������
( �)5��
n�5�5"$:�1��/9%��3& �63,'
n *2;3 .�-+6=><7��4801�39'
n���3 �• 1. #�3�5��• 2. #�3��=><5��• 3. AI5�!�• 4. ����489 • …
16
���(� � – 1n �.0-*'*����/��)��#$"��,� & %��!+���
n �1���*��*�AI� *�
17
����������#"���������� ��������$��������!����� ��
Why Should I Trust You?: Explaining the Predictions of Any Classifier, KDD’16.
���*� – 2n �),��463 ��5��+��%&$��2�("'��#/��• �/��463("′, %′) ���&$�("$.�463",��-)0!.��1/�8
n�7 ���,�
• b
18
��������� ���� ����
�������
���� �������������
Understanding Black-box Predictions via Influence Functions, ICML’17.
���,��� – 3n ��9��-58321 #��,��*��)0$ (+*AI-��+)0��
n �:!��-�-�AI" -��• ��,AI1��&�/0��,647.�' %0
19
���� whenRelationship ≠ Not-in-family, WifeCapital Gain < 7370
���� whenRelationship ≠ Not-in-familyCapital Gain >= 7370
���� whenRelationship ≠ Not-in-family, UnmarriedCapital Gain < 5095Capital Loss < 2114
���� whenRelationship = Not-in-familyCountry ≠ China, PeruCapital Gain < 5095
���� whenRelationship ≠ Not-in-familyCountry ≠ ChinaCapital Gain < 5095
���� whenRelationship ≠ Not-in-familyCapital Gain < 7370
Making Tree Ensembles Interpretable: A Bayesian Model Selection Approach, AISTATS’18.
���(� – 4n �.0-*'* ����/��)��"#!��,����%��&�$�� +��
n�1����*�
20Generating Visual Explanations, ECCV’16.
������
n �AI�����������
n �AI���������
n �AI������������
21
AI97��8��->0)Dn 2016� !�#5$AI@��/=��%*��->0"
n .).! AI97��*B5+4;� �68C��->0<'68�(5&"
n:.?!B�7��38C$AI@��/=��%61&27��A����@�,=��*�(0"
22
�����������
AI96��74-��*<1$4$%Cn� A �6��&,<1;"����27����5%01$;"
n� B :.���6��&'/>3�:01$4$"#37��+1$<8!� &��+1(<;"
n� C ��6� 3�6�6@B?&?A1$;")6?A=��+4$3!��7��*<4$"
23
�����B, C�9��
��B; � �3�&�4./$2$�n���+(��
• � 2���03!����2� �"&��)8/$7�• 13��9�:-6���3��&��0'7%<
n�8,� �&�58/*4./$7��5#7�
24
!$� #"���������������� �����������������
���CQ ��1�5�5GNC'CL0$<
n �����5� 5(6?@EI?&; • WHI’18Q ICML’185MNABJDH
� � 14/17 '?@EI?&;
� ��&;6Google Brain, IBM Research, Microsoft
• XAI’18Q ICML’185MNABJDH� � 26/30 '?@EI?&;
� ��&;6Samsung Research, Spotify, OTTO , ��ORP
n ��5��5(6?@EI?'KNF • ���'!)>3��#.,;��+93$R"1$%��4/$0���5����=�70$<)1'$
• ,-*Google Brain:IBM32������8���
25
���������������������� ���������
����������� � ���
26
�����
��
� &%�'!��#
���$%�'!��#
����
AI��" ������
�������������
���������� ������
27
�� "�
.�1
���076���������59 �
���
���+(����07����*�������0738�
/#�2,�$&
!���2,
�������� ������
����:4���%�2,-�����
)'38���������
:`e\O!�;QOf�O�"Mg��
n.�O8b[_adZ]X[49@JATK�E<7• 8CVM��9NP68CVM.��9U�=K*<7• 8CVM��9O� 6���B@�,7
nWY^cW"�EDh$%O��HB• �+@1SIM<OPLVM��?i• �+KEI�U�(DGTRA?i
n�O�"�EDh��-�QO&2O� • )�3@�>I<T/5P�?i• #O�O/5N�EI6LVM'U0FTRA?i
28
������� ��� ����� �����
7 �#K$�8QCN�Y�F
n:?KV1��KJM!�L3_a]M"�4Y�/AGS1�!N.A:2
n QCN15*��M��>�%ED��#K�6Y�FEG1�M�L"�AD:2• AIPM� IAD��Y��#K�0L&IA,R@IH1�'#L(�H?Wb=SAXK:c@IY1�M�L)+AGSU;2
• � IAD��MQQ�<@ZH:GS1�M�N�L-QK:2
n 5�%ED�6>9XO1[\^`[S!�S�?TB:NC2
29
���� ���������!� ��"�
n����AI�����• %#&$&�
Sun-In Lee ����
30
we developed and tested a machine learning based system called Prescience that predicts
real-time hypoxemia risk and presents an explanation of factors contributing to that risk
Prescience improved anesthesiologists' performance when providing interpretable hypoxemia risks with contributing factors.
The results suggest that if anesthesiologists currently anticipate 15% of events, then with Prescience assistance they could anticipate
30% of events
https://www.biorxiv.org/content/early/2017/10/21/206540
���������������������
n� ��AI����• DeepMind
31
� ����������������"��� ������
AI������!������
https://www.nature.com/articles/s41591-018-0107-6
WQY
n('9;AI_5!I] #<T,-U����S?]D9+�*SkmcT��T4$VU/DLO@R@:
n,-D�T�TgmdQdlO@]�1�D?]:• abfjaT,-0T;F`R5!?LK\�H[R@n<Q@A�5S�NE5!#T,-D�@:
n(�C\T;gmdT)�<D��:• %&QGO��_9��*R�8S2QG7XFQP9�3*S4"�1CZG^R@:
• ��TWW�B7`PZ96ZheimSR\R@:
n WJU9=5!#T� D�.LK��*R�>_�M�LO9�T�S)�GK@:
32
#� – 1d 8AIQ2�9R�,P5Z[_
n3�NEL'-DXL;W* R68<SA;KI��9JB?��DXL;W�+�?:W7
n2�"��PR6�G��QAI/^c]M!0?�.7• %$Q2�"R��EP�=WTQMRO;7
n2�PR/)b\c[T�.7• HXOVP/)Z[_?>>W�"?�;7
• ��PUKLR64�QAIP=L6Q2�&AIY�W�.T:W7
n “12�”T:V�W7• 2�Y��(Pa[bc`FWU<P^c]Y��M@WCN?� DXL;W7
33
�����
���
!� – 2X �(D�C*�D9OQ<DIF3/n+� R-�?C1AII��?C�5O2ETUWV?CL.1FSEF:��2 E34�)I6AN:@;GI#5F3/
n��G�?3HI1+�2DIF:.1'�2M1��2/
n AI8�0F�,D"%>P.AH��8�G&>PO=ED.1FSEF:��2H�I$BC3:7L?PF3/• 1JSF8�BC3O7N ��2→ 1��2/• 1AIH+�2H��I.1JSF8�BC3O7N2H��H��R�KO=E/
34