Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A SAT-centered XAI method for Deep Learning based Video Understanding

Hojer Key

TL;DR

The paper addresses the need for formal, verifiable explanations in deep video understanding by proposing a SAT-centered XAI framework that encodes a neural video model and video data into a CNF representation. It introduces a conceptual framework, formal encoding methods, and SAT-based queries for abductive ('Why?') and contrastive ('Why not?') explanations, instantiated in a novel architecture that couples a SAT solver with a video-understanding model. The key contributions include formal definitions, encoding techniques for spatio-temporal data, algorithms for explanation generation, and sketches of soundness and completeness proofs, aimed at offering explanations with formal guarantees. The work highlights the potential for rigorous, verifiable explanations in high-stakes video domains while acknowledging computational scalability and representational challenges, outlining a path for future extensions and human-readable outputs.

Abstract

This paper introduces a novel formal SAT-based explanation model for deep learning in video understanding. The proposed method integrates SAT solving techniques with the principles of formal explainable AI to address the limitations of existing XAI techniques in this domain. By encoding deep learning models and video data into a logical framework and formulating explanation queries as satisfiability problems, the method aims to generate logic-based explanations with formal guarantees. The paper details the conceptual framework, the process of encoding deep learning models and video data, the formulation of "Why?" and "Why not?" questions, and a novel architecture integrating a SAT solver with a deep learning video understanding model. While challenges related to computational complexity and the representational power of propositional logic remain, the proposed approach offers a promising direction for enhancing the explainability of deep learning in the complex and critical domain of video understanding.

A SAT-centered XAI method for Deep Learning based Video Understanding

TL;DR

The paper addresses the need for formal, verifiable explanations in deep video understanding by proposing a SAT-centered XAI framework that encodes a neural video model and video data into a CNF representation. It introduces a conceptual framework, formal encoding methods, and SAT-based queries for abductive ('Why?') and contrastive ('Why not?') explanations, instantiated in a novel architecture that couples a SAT solver with a video-understanding model. The key contributions include formal definitions, encoding techniques for spatio-temporal data, algorithms for explanation generation, and sketches of soundness and completeness proofs, aimed at offering explanations with formal guarantees. The work highlights the potential for rigorous, verifiable explanations in high-stakes video domains while acknowledging computational scalability and representational challenges, outlining a path for future extensions and human-readable outputs.

Abstract

This paper introduces a novel formal SAT-based explanation model for deep learning in video understanding. The proposed method integrates SAT solving techniques with the principles of formal explainable AI to address the limitations of existing XAI techniques in this domain. By encoding deep learning models and video data into a logical framework and formulating explanation queries as satisfiability problems, the method aims to generate logic-based explanations with formal guarantees. The paper details the conceptual framework, the process of encoding deep learning models and video data, the formulation of "Why?" and "Why not?" questions, and a novel architecture integrating a SAT solver with a deep learning video understanding model. While challenges related to computational complexity and the representational power of propositional logic remain, the proposed approach offers a promising direction for enhancing the explainability of deep learning in the complex and critical domain of video understanding.

Paper Structure

This paper contains 47 sections, 3 equations, 1 figure, 2 algorithms.

Table of Contents

  1. Introduction and Background
  2. Formal Explainable Artificial Intelligence (XAI): Theoretical Underpinnings
  3. Logic-Based Approaches to Explainability
  4. Abductive and Contrastive Explanations
  5. Formal Verification of AI Systems
  6. Related Work
  7. Combining SAT Solving and Formal XAI for Machine Learning Explanation
  8. Using SAT for Explaining Classifiers and Other ML Models
  9. Logic-Based Rule Extraction from Machine Learning Models
  10. Explainable AI for Deep Learning Models
  11. Post-hoc Explanation Techniques for Neural Networks
  12. Limitations of Existing Methods in Providing Formal Guarantees
  13. Explainable AI in the Domain of Video Understanding
  14. Current Techniques and Their Shortcomings
  15. Challenges Specific to Explaining Spatio-Temporal Models
  16. ...and 32 more sections

Figures (1)

  • Figure 1: High-Level Architecture of the Proposed SAT-Based Explanation Model for Deep Video Understanding