Towards Automatic Concept-based Explanations

Abstract

Interpretability has become an important topic of research as more machine learning (ML) models are deployed and widely used to make important decisions. Most of the current explanation methods provide explanations through feature importance scores, which identify features that are important for each individual input. However, how to systematically summarize and interpret such per sample feature importance scores itself is challenging. In this work, we propose principles and desiderata for \emph{concept} based explanation, which goes beyond per-sample features to identify higher-level human-understandable concepts that apply across the entire dataset. We develop a new algorithm, ACE, to automatically extract visual concepts. Our systematic experiments demonstrate that \alg discovers concepts that are human-meaningful, coherent and important for the neural network's predictions.

References in corpus (14)

• Towards A Rigorous Science of Interpretable Machine Learning
• Axiomatic Attribution for Deep Networks
• European Union regulations on algorithmic decision-making and a "right to explanation"
• Understanding Black-box Predictions via Influence Functions
• ParseNet: Looking Wider to See Better
• SmoothGrad: removing noise by adding noise
• Deep Image Prior
• Model-Agnostic Interpretability of Machine Learning
• Sanity Checks for Saliency Maps
• The (Un)reliability of saliency methods
• Interpretation of Neural Networks is Fragile
• Methods and Models for Interpretable Linear Classification
• Knockoffs for the mass: new feature importance statistics with false discovery guarantees
• Multiresolution hierarchy co-clustering for semantic segmentation in sequences with small variations

Cited by in corpus (12)

• Post-hoc Interpretability for Neural NLP: A Survey
• Acquisition of Chess Knowledge in AlphaZero
• Logic Explained Networks
• Entropy-based Logic Explanations of Neural Networks
• Towards Relatable Explainable AI with the Perceptual Process
• Interpretability of a Deep Learning Model in the Application of Cardiac MRI Segmentation with an ACDC Challenge Dataset
• Model-Based Counterfactual Synthesizer for Interpretation
• Explaining in Style: Training a GAN to explain a classifier in StyleSpace
• From Heatmaps to Structural Explanations of Image Classifiers
• Towards Robust Metrics for Concept Representation Evaluation
• PyTorch, Explain! A Python library for Logic Explained Networks
• Example-Based Concept Analysis Framework for Deep Weather Forecast Models