tutorial on interpreting and explaining deep models in
play

Tutorial on Interpreting and Explaining Deep Models in Computer - PowerPoint PPT Presentation

Nov 16, 2023 •291 likes •630 views

Tutorial on Interpreting and Explaining Deep Models in Computer Vision Wojciech Samek Grgoire Montavon Klaus-Robert Mller (Fraunhofer HHI) (TU Berlin) (TU Berlin) 08:30 - 09:15 Introduction KRM 09:15 - 10:00 Techniques for Interpretability

  1. Tutorial on Interpreting and Explaining Deep Models in Computer Vision Wojciech Samek Grégoire Montavon Klaus-Robert Müller (Fraunhofer HHI) (TU Berlin) (TU Berlin) 08:30 - 09:15 Introduction KRM 09:15 - 10:00 Techniques for Interpretability GM 10:00 - 10:30 Coffee Break ALL 10:30 - 11:15 Applications of Interpretability WS 11:15 - 12:00 Further Applications and Wrap-Up KRM

  2. O v e r v i e w o f E x p l a n a t i o n M e t h o d s Q u e s t i o n : W h i c h o n e t o c h o o s e ? 2 / 3 3

  3. F i r s t A t t e mp t : D i s t a n c e t o G r o u n d T r u t h i n p u t e v i d e n c e f o r D N N “ t r u c k ” e x p l a n a t i o n e r r o r g r o u n d ? ? ? t r u t h 3 / 3 3

  4. F i r s t A t t e mp t : D i s t a n c e t o G r o u n d T r u t h i n p u t e v i d e n c e f o r D N N “ c a r ” e x p l a n a t i o n e r r o r g r o u n d ? ? ? t r u t h ? ? ? 4 / 3 3

  5. F r o m G r o u n d T r u t h E x p l a n a t i o n s t o A x i o ms I d e a : E v a l u a t e t h e e x p l a n a t i o n t e c h n i q u e a x i o m a t i c a l l y , i . e . i t m u s t p a s s a n u m b e r o f p r e d e fi n e d “ u n i t t e s t s ” . [ S u n ’ 1 1 , B a c h ’ 1 5 , M o n t a v o n ’ 1 7 , S a m e k ’ 1 7 , S u n d a r a j a n ’ 1 7 , K i n d e r m a n s ’ 1 7 , M o n t a v o n ’ 1 8 ] . e x p l a n a t i o n t e c h n i q u e D N N 5 / 3 3

  6. P r o p e r t i e s 1 - 2 : C o n s e r v a t i o n a n d P o s i t i v i t y [ M o n t a v o n ’ 1 7 , s e e a l s o S u n ’ 1 1 , L a n d e c k e r ’ 1 3 , B a c h ’ 1 5 ] e x p l a n a t i o n D N N C o n s e r v a t i o n : T o t a l a t t r i b u t i o n o n t h e i n p u t f e a t u r e s s h o u l d b e p r o p o r t i o n a l t o t h e a m o u n t o f ( e x p l a i n a b l e ) e v i d e n c e a t t h e o u t p u t . P o s i t i v i t y : I f t h e n e u r a l n e t w o r k i s c e r t a i n a b o u t i t s p r e d i c t i o n , i n p u t f e a t u r e s a r e e i t h e r r e l e v a n t ( p o s i t i v e ) o r i r r e l e v a n t ( z e r o ) . 6 / 3 3

  7. P r o p e r t y 3 : C o n t i n u i t y [ M o n t a v o n ’ 1 8 ] I f t w o i n p u t s a r e t h e a l m o s t t h e s a m e , a n d t h e p r e d i c t i o n i s a l s o a l m o s t t h e s a m e , t h e n t h e e x p l a n a t i o n s h o u l d a l s o b e a l m o s t t h e s a m e . E x a mp l e : M e t h o d 1 M e t h o d 2 7 / 3 3

  8. T e s t i n g C o n t i n u i t y e x p l a n a t i o n i n p u t s c o r e s f ( x ) L R P - α β 1 0 R i S e n s i t i v i t y a n a l y s i s 8 / 3 3

  9. P r o p e r t y 4 : S e l e c t i v i t y [ B a c h ’ 1 5 , S a m e k ’ 1 7 ] M o d e l m u s t a g r e e w i t h t h e e x p l a n a t i o n : I f i n p u t f e a t u r e s a r e a t t r i b u t e d r e l e v a n c e , r e m o v i n g t h e m s h o u l d r e d u c e e v i d e n c e a t t h e o u t p u t . E x a mp l e : M e t h o d 1 M e t h o d 2 9 / 3 3

  10. T e s t i n g S e l e c t i v i t y w i t h P i x e l - F l i p p i n g [ B a c h ’ 1 5 , S a m e k ’ 1 7 ] e x p l a n a t i o n i n p u t s c o r e s f ( x ) L R P - α β 1 0 ( ) f x S e n s i t i v i t y a n a l y s i s 1 0 / 3 3

  11. t u t p u n p I n I x ) 2 P x ) 2 P t B t B n β n d m e d i e 0 r e i e α d o d - d 1 f i d a i i P a r u r u n R G G . U ( G G L ( . E x p l a n a t i o n t e c h n i q u e s . P r o p e r t i e s ✔ ✔ ✔ ✔ 1 . C o n s e r v a t i o n ✔ ✔ ✔ ✔ ✔ 2 . P o s i t i v i t y ✔ ✔ ✔ ✔ 3 . C o n t i n u i t y ✔ ✔ ✔ ✔ ✔ 4 . S e l e c t i v i t y . . . 1 1 / 3 3

  12. Q u e s t i o n : C a n w e d e d u c e s o m e p r o p e r t i e s w i t h o u t e x p e r i m e n t s , d i r e c t l y f r o m t h e e q u a t i o n s ? 1 2 / 3 3

  13. R e mi n d e r B a c k p r o p i n t e r n a l s ( f o r p r o p a g a t i n g g r a d i e n t ) L R P - i n t e r n a l s ( f o r p r o p a g a t i n g r e l e v a n c e ) α β 1 0 1 3 / 3 3

  14. E x a mp l e : D e d u c i n g C o n s e r v a t i o n v s . g r a d i n p u t × L R P - p r o p a g a t i o n r u l e α β 1 0 × i n p u t S u m m i n g g i v e s t h e p r o p e r t y W h e n b i a s i s n e g a t i v e , g r a d × i n p u t w i l l t e n d t o i n fl a t e s c o r e s . 1 4 / 3 3

  15. E x a mp l e : D e d u c i n g C o n t i n u i t y L R P - p r o p a g a t i o n r u l e α β v s . g r a d i n p u t × 1 0 ( w h e n b i a s n e g a t i v e , c o n t i n u i t y d u e t o d e n o m i n a t o r u p p e r - b o u n d i n g n u m e r a t o r . ) 1 5 / 3 3

  16. I n t e r me d i a t e C o n c l u s i o n G r o u n d - t r u t h e x p l a n a t i o n s a r e e l u s i v e . I n p r a c t i c e , w e a r e r e d u c e d t o v i s u a l a s s e s s m e n t o r t o t e s t t h e e x p l a n a t i o n f o r a n u m b e r o f a x i o m s . S o m e p r o p e r t i e s c a n b e d e d u c e d f r o m t h e s t r u c t u r e o f t h e e x p l a n a t i o n m e t h o d . O t h e r c a n b e t e s t e d e m p i r i c a l l y . α 1 β 0 - L R P s a t i s fi e s k e y p r o p e r t i e s o f a n e x p l a n a t i o n . S e n s i t i v i t y a n a l y s i s a n d g r a d i e n t i n p u t h a v e × c r u c i a l l i m i t a t i o n s . 1 6 / 3 3

  17. F r o m L R P t o D e e p T a y l o r D e c o mp o s i t i o n T h e L R P - α β r u l e 1 0 c a n b e s e e n a s a d e e p T a y l o r [ M o n t a v o n ’ 1 7 ] d e c o m p o s i t i o n ( D T D ) w h i c h t h e n y i e l d s d o m a i n - a n d l a y e r - s p e c i fi c r u l e s 1 7 / 3 3

  18. D T D : T h e S t r u c t u r e o f R e l e v a n c e P r o p o s i t i o n : R e l e v a n c e a t e a c h l a y e r i s a p r o d u c t o f t h e a c t i v a t i o n a n d a n a p p r o x i m a t e l y c o n s t a n t t e r m . 1 8 / 3 3

  19. D T D : T h e R e l e v a n c e a s a N e u r o n 1 9 / 3 3

  20. D T D : T a y l o r E x p a n s i o n o f t h e R e l e v a n c e 2 0 / 3 3

  21. D T D : D e c o mp o s i n g t h e R e l e v a n c e T a y l o r e x p a n s i o n a t r o o t p o i n t : R e l e v a n c e c a n n o w b e b a c k w a r d p r o p a g a t e d 2 1 / 3 3

  22. D T D : C h o o s i n g t h e R o o t P o i n t ( D e e p T a y l o r g e n e r i c ) C h o i c e o f r o o t p o i n t 1 . n e a r e s t r o o t ✔ ✔ 2 . r e s c a l e d a c t i v a t i o n 3 . r e s c a l e d e x c i t a t i o n s ✔ ✔ ( L R P - ) α β 1 0 2 2 / 3 3

  23. D T D : V e r i f y i n g t h e P r o d u c t S t r u c t u r e B u t w a s i t t r u e ? 1 . a s s u m e i t h o l d s i n h i g h e r - l a y e r 2 . a p p l y L R P - r u l e α β 1 0 3 . o b s e r v e i t a l s o h o l d s i n l o w e r - l a y e r 2 3 / 3 3

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

Tutorial on Interpreting and Explaining Deep Models in Computer Vision Wojciech Samek Grgoire

Tutorial on Interpreting and Explaining Deep Models in Computer Vision Wojciech Samek Grgoire

Tutorial on Interpreting and Explaining Deep Models in Computer Vision Wojciech Samek Grgoire Montavon Klaus-Robert Mller (Fraunhofer HHI) (TU Berlin) (TU Berlin) 08:30 - 09:15 Introduction KRM 09:15 - 10:00 Techniques for Interpretability

676 views • 63 slides
Tutorial on Interpreting and Explaining Deep Models in Computer Vision Wojciech Samek Grgoire

Tutorial on Interpreting and Explaining Deep Models in Computer Vision Wojciech Samek Grgoire

Tutorial on Interpreting and Explaining Deep Models in Computer Vision Wojciech Samek Grgoire Montavon Klaus-Robert Mller (Fraunhofer HHI) (TU Berlin) (TU Berlin) 08:30 - 09:15 Introduction KRM 09:15 - 10:00 Techniques for Interpretability

416 views • 23 slides
Tutorial on Methods for Interpreting and Understanding Deep Neural Networks Wojciech Samek

Tutorial on Methods for Interpreting and Understanding Deep Neural Networks Wojciech Samek

Tutorial on Methods for Interpreting and Understanding Deep Neural Networks Wojciech Samek Grgoire Montavon Klaus-Robert Mller (Fraunhofer HHI) (TU Berlin) (TU Berlin) 1:30 - 2:00 Part 1: Introduction 2:00 - 3:00 Part 2a: Making Deep

431 views • 29 slides
ICASSP 2017 Tutorial on Methods for Interpreting and Understanding Deep Neural Networks G.

ICASSP 2017 Tutorial on Methods for Interpreting and Understanding Deep Neural Networks G.

ICASSP 2017 Tutorial on Methods for Interpreting and Understanding Deep Neural Networks G. Montavon, W. Samek, K.-R. Mller Part 2: Making Deep Neural Networks Transparent 5 March 2017 Making Deep Neural Nets Transparent DNN transparency

448 views • 44 slides
From ML Successes to Applications ICIP18 Tutorial on Interpretable Deep Learning 2 Black Box

From ML Successes to Applications ICIP18 Tutorial on Interpretable Deep Learning 2 Black Box

From ML Successes to Applications ICIP18 Tutorial on Interpretable Deep Learning 2 Black Box Models Huge volumes of data Solve task Computing power Deep Neural Network Information (implicit) ICIP18 Tutorial on Interpretable Deep

540 views • 33 slides
A Tutorial on Deep Probabilistic Generative Models Ryan P. Adams Princeton University Machine

A Tutorial on Deep Probabilistic Generative Models Ryan P. Adams Princeton University Machine

A Tutorial on Deep Probabilistic Generative Models Ryan P. Adams Princeton University Machine Learning Summer School Buenos Aires, Argentina June 2018 lips.cs.princeton.edu @ryan_p_adams Tutorial Outline What is generative modeling?

2.33k views • 156 slides
4 Deep Generative Models BVM 2018 Tutorial: Advanced Deep Learning Methods Jens Petersen Dept.

4 Deep Generative Models BVM 2018 Tutorial: Advanced Deep Learning Methods Jens Petersen Dept.

4 Deep Generative Models BVM 2018 Tutorial: Advanced Deep Learning Methods Jens Petersen Dept. of Neuroradiology, Heidelberg University Hospital Div. of Medical Image Computing, DKFZ Heidelberg Faculty of Physics & Astronomy, Heidelberg

1.48k views • 65 slides
outline of this tutorial motivations 1 ACISS09 tutorial on deep belief nets deep

outline of this tutorial motivations 1 ACISS09 tutorial on deep belief nets deep

outline of this tutorial motivations 1 ACISS09 tutorial on deep belief nets deep autoencoders deep belief nets sigmoid belief nets 2 Marcus Frean why are they hard to train? could layer-by-layer training work? Melbourne, 2009

509 views • 15 slides
Visualizing and Interpreting Deep Neural Networks Bolei Zhou Department of Information

Visualizing and Interpreting Deep Neural Networks Bolei Zhou Department of Information

Visualizing and Interpreting Deep Neural Networks Bolei Zhou Department of Information Engineering The Chinese University of Hong Kong Deep Neural Networks are Everywhere Playing Go Making Medical Decision Understanding Scenes Deep Neural

1.1k views • 46 slides
UCL Tutorial on: Deep Belief Nets (An updated and extended version of my 2007 NIPS tutorial)

UCL Tutorial on: Deep Belief Nets (An updated and extended version of my 2007 NIPS tutorial)

UCL Tutorial on: Deep Belief Nets (An updated and extended version of my 2007 NIPS tutorial) Geoffrey Hinton Canadian Institute for Advanced Research & Department of Computer Science University of Toronto Schedule for the Tutorial

1.66k views • 126 slides
Explaining Deep Learning Predictions and Isaac Ahern Integrating Domain Ontologies Outline

Explaining Deep Learning Predictions and Isaac Ahern Integrating Domain Ontologies Outline

Explaining Deep Learning Predictions and Integrating Domain Ontologies Explaining Deep Learning Predictions and Isaac Ahern Integrating Domain Ontologies Outline Project Background problems Isaac Ahern domains Explaining any

646 views • 35 slides
Deep Learning Tutorial Part I Greg Shakhnarovich TTI-Chicago December 2016 Deep Learning

Deep Learning Tutorial Part I Greg Shakhnarovich TTI-Chicago December 2016 Deep Learning

Deep Learning Tutorial Part I Greg Shakhnarovich TTI-Chicago December 2016 Deep Learning Tutorial,Part I 1 Overview Goals of the tutorial Somewhat organized overview of basics, and some more advanced topics Demistify jargon Pointers for

1.25k views • 105 slides
Deep Learning Tutorial Part II Greg Shakhnarovich TTI-Chicago December 2016 Deep Learning

Deep Learning Tutorial Part II Greg Shakhnarovich TTI-Chicago December 2016 Deep Learning

Deep Learning Tutorial Part II Greg Shakhnarovich TTI-Chicago December 2016 Deep Learning Tutorial,Part II 1 Overview Goals of the tutorial Somewhat organized overview of basics, and some more advanced topics Demistify jargon Pointers

925 views • 63 slides
Interpreting Models for Categorical and Count Outcomes Rose Medeiros StataCorp LLC Stata

Interpreting Models for Categorical and Count Outcomes Rose Medeiros StataCorp LLC Stata

Introduction Estimation Postestimation Conclusion Interpreting Models for Categorical and Count Outcomes Rose Medeiros StataCorp LLC Stata Webinar March 21, 2019 Interpreting Models for Categorical and Count Outcomes Handout page: 1

559 views • 54 slides
Interpreting Embedding Models of Knowledge Bases: Model Agnostic Approaches 2018 ICML Workshop on

Interpreting Embedding Models of Knowledge Bases: Model Agnostic Approaches 2018 ICML Workshop on

Introduction & Background Interpreting Embedding Models of KBs Experiments Conclusion Interpreting Embedding Models of Knowledge Bases: Model Agnostic Approaches 2018 ICML Workshop on Human Interpretability in Machine Learning Arthur C.

751 views • 24 slides
Why Did You Say That? Explaining and Diversifying Captioning Models Kate Saenko VQA Workshop,

Why Did You Say That? Explaining and Diversifying Captioning Models Kate Saenko VQA Workshop,

Why Did You Say That? Explaining and Diversifying Captioning Models Kate Saenko VQA Workshop, CVPR, July 26, 2017 Explaining: Top-down saliency guided by captions http://ai.bu.edu/caption-guided-saliency/ Vasili Abir Jianming Kate

726 views • 53 slides
Modelling XML Applications Patryk Czarnik XML and Applications 2013/2014 Lecture 2

Modelling XML Applications Patryk Czarnik XML and Applications 2013/2014 Lecture 2

Modelling XML Applications Patryk Czarnik XML and Applications 2013/2014 Lecture 2 14.10.2013 XML application (recall) XML application ( zastosowanie XML ) A concrete language with XML syntax T ypically defjned as: Fixed set of

490 views • 30 slides
Repair of XML documents (w.r.t. given DTD) Robert Surwka 1 Outline of the

Repair of XML documents (w.r.t. given DTD) Robert Surwka 1 Outline of the

Repair of XML documents (w.r.t. given DTD) Robert Surwka 1 Outline of the problem Convert given xml such that it

958 views • 27 slides
Outline XML Documents: Quick Review on XML Limitations and Opportunities The Problem

Outline XML Documents: Quick Review on XML Limitations and Opportunities The Problem

Relational Databases for Querying Outline XML Documents: Quick Review on XML Limitations and Opportunities The Problem Definition General Approach to Querying XML using a Jayavel Shanmugasundaram RDBMS Kristin Tufte Gang He

777 views • 6 slides
Kernel Modules KLD Device Drivers The kld interface allows system administrators to

Kernel Modules KLD Device Drivers The kld interface allows system administrators to

Dynamic Kernel Linker Facility - Kernel Modules KLD Device Drivers The kld interface allows system administrators to dynamically add and remove functionality from a Pseudo device example running system. Joystick device example This

503 views • 8 slides
Window System API Agenda 1. Terminology 2. TopComponent 3. WindowManager 4. Mode

Window System API Agenda 1. Terminology 2. TopComponent 3. WindowManager 4. Mode

Window System API Agenda 1. Terminology 2. TopComponent 3. WindowManager 4. Mode 5. TopComponentGroup 6. Configuration Options 7. Case Studies 1. Terminology What is a window system? A window system is a system for

462 views • 44 slides
Introduction to HTML5 Where to start learning about HTML5? HTML 5 differences from HTML 4

Introduction to HTML5 Where to start learning about HTML5? HTML 5 differences from HTML 4

Introduction to HTML5 Where to start learning about HTML5? HTML 5 differences from HTML 4 http://w3.org/TR/html5-diff/ HTML 5 HTML 4 http://standards.mitsue.co.jp/ resources/w3c/TR/html5-diff/ HTML5 differences

684 views • 35 slides
When are we committed to crossing critical (1.5 or 2 C) temperature thresholds? Cristian

When are we committed to crossing critical (1.5 or 2 C) temperature thresholds? Cristian

When are we committed to crossing critical (1.5 or 2 C) temperature thresholds? Cristian Proistosescu 1 Kyle Armour 1 Gerard Roe 1 Peter Huybers 2 1 University of Washington 2 Harvard University AGU Fall Meeting 2017 Courtesy of NASAs Earth

435 views • 29 slides
3. Defining the document structure (DTD) Declaration of application-specific names and

3. Defining the document structure (DTD) Declaration of application-specific names and

3. Defining the document structure (DTD) Declaration of application-specific names and structural constraints A document is valid if it specifies a DTD, and if its contents conform to the DTD. A validating parser does the checking;

895 views • 18 slides

More recommend