Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)
read the original abstract
The interpretation of deep learning models is a challenge due to their size, complexity, and often opaque internal state. In addition, many systems, such as image classifiers, operate on low-level features rather than high-level concepts. To address these challenges, we introduce Concept Activation Vectors (CAVs), which provide an interpretation of a neural net's internal state in terms of human-friendly concepts. The key idea is to view the high-dimensional internal state of a neural net as an aid, not an obstacle. We show how to use CAVs as part of a technique, Testing with CAVs (TCAV), that uses directional derivatives to quantify the degree to which a user-defined concept is important to a classification result--for example, how sensitive a prediction of "zebra" is to the presence of stripes. Using the domain of image classification as a testing ground, we describe how CAVs may be used to explore hypotheses and generate insights for a standard image classification network as well as a medical application.
This paper has not been read by Pith yet.
Forward citations
Cited by 11 Pith papers
-
When Are Two Networks the Same? Tensor Similarity for Mechanistic Interpretability
Tensor similarity is a symmetry-invariant metric that measures functional equivalence between tensor-based networks using a recursive algorithm for cross-layer mechanisms.
-
From Mechanistic to Compositional Interpretability
Compositional interpretability defines explanations as commuting syntactic-semantic mapping pairs grounded in compositionality and minimum description length, with compressive refinement and a parsimony theorem guaran...
-
Unifying machine learning and quantum chemistry -- a deep neural network for molecular wavefunctions
Deep neural network predicts molecular wavefunctions in atomic orbital basis from which quantum properties are derived at force-field efficiency.
-
Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces
A latent mediation framework with sparse autoencoders enables non-additive token-level influence attribution in LLMs by learning orthogonal features and back-propagating attributions.
-
FeatMap: Understanding image manipulation in the feature space and its implications for feature space geometry
Linear mappings in feature space can reconstruct a wide range of image manipulations including semantic edits, suggesting that feature representations are approximately linearly organized.
-
Finding Meaning in Embeddings: Concept Separation Curves
Concept Separation Curves provide a classifier-independent method to visualize and quantify how sentence embeddings distinguish conceptual meaning from syntactic variations across languages and domains.
-
UNBOX: Unveiling Black-box visual models with Natural-language
UNBOX recovers interpretable text concepts that maximally activate classes in black-box vision models by recasting activation maximization as semantic search with LLMs and diffusion models.
-
CLIF: Concept-Level Influence Functions for Transparent Bottleneck Models
CLIF applies influence functions to pinpoint influential training samples and key concepts in Concept Bottleneck Models, enabling data debugging and behavioral insights on CEBaB and Yelp datasets.
-
Generative Counterfactual Introspection for Explainable Deep Learning
A generative-model-driven introspection method produces counterfactual image edits to explain deep neural network predictions on MNIST and CelebA.
-
Explainable Artificial Intelligence Techniques for Interpretation of Food Models: a Review
A survey proposing a taxonomy of XAI techniques for food quality research organized by data types and explanation methods.
-
Unexplainability and Incomprehensibility of Artificial Intelligence
Advanced AI systems are unexplainable in full and produce explanations that humans cannot comprehend.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.