SLIDE 1 Funct Functiona nal Trans nspa parency ncy for r Struct ructur ured d Data: a a Gam ame-The heoretic c Appr pproach ch
Gu Guang-He He Lee, Lee, We Wengong Jin Jin, , David d Alvarez Me Melis, , and nd Tom
Jaakkola
SLIDE 2 Goal: al: understan and the (comple lex) ne network
dog
de deep p ne nets
Img: https://blog.paperspace.com/intro-to-optimization-in-deep-learning-gradient-descent/
SLIDE 3 De Deep p ne nets
Img: https://blog.paperspace.com/intro-to-optimization-in-deep-learning-gradient-descent/
Typic ical al method: post-hoc explan lanatio ion
SLIDE 4 De Deep p ne nets
Img: https://blog.paperspace.com/intro-to-optimization-in-deep-learning-gradient-descent/
Typic ical al method: post-hoc explan lanatio ion
- 1. Given an example
- 2. choose a neighborhood
!
SLIDE 5 De Deep p ne nets
Img: https://blog.paperspace.com/intro-to-optimization-in-deep-learning-gradient-descent/
Typic ical al method: post-hoc explan lanatio ion
- 1. Given an example
- 3. Find a simple approximation
- e.g., linear model, decision tree.
≈
SLIDE 6 De Deep p ne nets
Img: https://blog.paperspace.com/intro-to-optimization-in-deep-learning-gradient-descent/
Typic ical al method: post-hoc explan lanatio ion
- 1. Given an example
- 3. Find a simple approximation
- e.g., linear model, decision tree.
≈
SLIDE 7 Po Post-hoc explan lanatio ions ar are not stab able le
Input 1 Explanation 1 (Alvarez-Melis & Jaakkola, 18’, Ghorbani et al., 19’)
SLIDE 8 Po Post-hoc explan lanatio ions ar are not stab able le
+ ! Input 1 Input 2 Explanation 1 Explanation 2 (Alvarez-Melis & Jaakkola, 18’, Ghorbani et al., 19’)
SLIDE 9
Reas ason: the network does not operate as as the desir ired explan lanatio ion
≈
SLIDE 10
Train ainin ing comple lex models ls to exhib ibit it mean anin ingful l propertie ies lo locally ally
stability, transparency, …
SLIDE 11
Train ainin ing comple lex models ls to exhib ibit it mean anin ingful l propertie ies lo locally ally
:the set of functions with desired property stability, transparency, …
SLIDE 12 Train ainin ing comple lex models ls to exhib ibit it mean anin ingful l propertie ies lo locally ally
stability, transparency, … :the set of functions with desired property
- example for transparency: linear model, decision tree
SLIDE 13
Train ainin ing comple lex models ls to exhib ibit it mean anin ingful l propertie ies lo locally ally
:the set of functions with desired property
Degree to which the property is enforced on around .
stability, transparency, …
SLIDE 14
Regularize the model towards the property
Train ainin ing comple lex models ls to exhib ibit it mean anin ingful l propertie ies lo locally ally
:the set of functions with desired property
Degree to which the property is enforced on around .
stability, transparency, …
SLIDE 15
For each , the witness measures the enforcement
Functio ional al property enforcement
SLIDE 16
For each , the witness measures the enforcement We regularize the predictor towards agreement
Functio ional al property enforcement
SLIDE 17
$
The asymmetry leads to efficiency in optimization. (see the paper for more details)
Functio ional al property enforcement
A co-operative game:
SLIDE 18
Exam ample les
Task Predictor Witness
SLIDE 19
Exam ample les
Task Predictor Witness
SLIDE 20 Exam ample les
Task Predictor Witness
toxic
(Hamilton et al., 17’)
SLIDE 21 Exam ample les
Task Predictor Witness
toxic
(Hamilton et al., 17’) (Jin et al., 18’)
SLIDE 22 Empir iric ical al study
n measur ure trans nspa parency ncy ba based d on n de devi viation n be between n pr predi dict ctor r and nd expl plaine ner.
SLIDE 23 Mo Mode dels traine ned d w/ thi his appr pproach ch yield d more compa pact ct expl plana nations ns
The explanation from
The explanation from a normal model
SLIDE 24 Poster: :
06:30 -- 09:00 PM @ Pacific Ballroom #64
- Details and analysis about the framework
Related w work on functional transparency:
Towards Robust, Locally Linear Deep Networks, ICLR 19’
