Contrastive Semantic Projection: Faithful Neuron Labeling with Contrastive Examples

TL;DR

This paper introduces Contrastive Semantic Projection (CSP), a novel method that uses contrastive examples with vision language models and CLIP encoders to produce more faithful and specific neuron labels in deep networks.

Contribution

It presents CSP, an extension of SemanticLens, integrating contrastive examples into neuron labeling to improve faithfulness and semantic detail.

Findings

Contrastive labeling yields more specific candidate labels.

CSP improves faithfulness and semantic granularity over baselines.

Experiments and case study demonstrate effectiveness in melanoma detection.

Abstract

Neuron labeling assigns textual descriptions to internal units of deep networks. Existing approaches typically rely on highly activating examples, often yielding broad or misleading labels by focusing on dominant but incidental visual factors. Prior work such as FALCON introduced contrastive examples -- inputs that are semantically similar to activating examples but elicit low activations -- to sharpen explanations, but it primarily addresses subspace-level interpretability rather than scalable neuron-level labeling. We revisit contrastive explanations for neuron-level labeling in two stages: (1) candidate label generation with vision language models (VLMs) and (2) label assignment with CLIP-like encoders. First, we show that providing contrastive image sets to VLMs yields candidate labels that are more specific and more faithful. Second, we introduce Contrastive Semantic Projection (CSP), an extension of SemanticLens that incorporates contrastive examples directly into its CLIP-based scoring and selection pipeline. Across extensive experiments and a case study on melanoma detection, contrastive labeling improves both faithfulness and semantic granularity over state-of-the-art baselines. Our results demonstrate that contrastive examples are a simple yet powerful and currently underutilized component of neuron labeling and analysis pipelines.