Overview
Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!
Explore methods, applications, and recent developments in Explainable AI with Dr. Wojciech Samek in this 43-minute conference talk from ODSC Europe 2019. Discover the effectiveness of explanation techniques like Layer-wise Relevance Propagation (LRP) across various data types and neural architectures. Learn how LRP provides insights into individual predictions through heatmaps visualizing pixel relevance in decision-making processes. Gain understanding of perturbation methods, Taylor expansions, and best practices for LRP implementation. Examine LRP applications in diverse problem domains, strategies for interpreting prediction strategies, model comprehension, and data analysis. Explore meta-explanations and Spectral Relevance Analysis (Spray) before concluding with a discussion on challenges and open questions in the field of explainable AI.
Syllabus
Intro
Explaining Predictions
Explanation Methods
Perturbation
Layer-wise Relevance Propagation
Approach 2: (Simple) Taylor Expansions
Simple Taylor Decomposition
Deep Taylor Decomposition
Best Practice for LRP
Evaluating Explanations
LRP Applied to Different Problems
Understanding Prediction Strategies
Understanding the Model
Understanding the Data
Meta-Explanations
Spectral Relevance Analysis (Spray)
Summary
Taught by
Open Data Science