{"slug": "spice-simple-polysemantic-feature-interpretation-via-clustering-based", "title": "SPICE: Simple Polysemantic Feature Interpretation via Clustering-based Explanation", "summary": "Researchers introduced SPICE (Simple Polysemantic Feature Interpretation via Clustering-based Explanation), a framework for analyzing polysemanticity in deep vision architectures, detailed in arXiv paper 2609.13198v1. SPICE avoids architecture-dependent propagation rules, enabling the first systematic comparison of polysemanticity across both CNNs and Transformers, and automatically determines the number of concept clusters per neuron, eliminating reliance on a preset K. The authors used SPICE to investigate how polysemanticity emerges, varies across depth and architecture, and forms through distinct computational pathways.", "body_md": "arXiv:2609.13198v1 Announce Type: new \nAbstract: One of the pivotal recent challenges in neural network interpretability is polysemanticity, where a single neuron is activated by multiple, often unrelated concepts, hindering clear functional understanding. Although prior work has explored this phenomenon, existing approaches remain architecture-specific and depend on manual heuristics such as a fixed number of concept clusters ($K$), limiting their generality and scalability--especially for modern Transformer-based models. To address these limitations, we introduce SPICE (\\textbf{S}imple \\textbf{P}olysemantic Feature \\textbf{I}nterpretation via \\textbf{C}lustering-based \\textbf{E}xplanation), a generalizable framework for analyzing polysemanticity in deep vision architectures. SPICE avoids architecture-dependent propagation rules, enabling the first systematic comparison of polysemanticity across both CNNs and Transformers, and automatically determines the number of concept clusters per neuron, eliminating reliance on a preset $K$ and supporting scalable analysis for large models. Using SPICE, we conduct a comprehensive investigation into how polysemanticity emerges, varies across depth and architecture, and forms through distinct computational pathways.", "url": "https://wpnews.pro/news/spice-simple-polysemantic-feature-interpretation-via-clustering-based", "canonical_source": "https://arxiv.org/abs/2609.13198", "published_at": "2026-09-15 04:00:00+00:00", "updated_at": "2026-09-15 04:30:15.906335+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "neural-networks", "computer-vision", "ai-research"], "entities": ["SPICE", "arXiv", "CNNs", "Transformers"], "alternates": {"html": "https://wpnews.pro/news/spice-simple-polysemantic-feature-interpretation-via-clustering-based", "markdown": "https://wpnews.pro/news/spice-simple-polysemantic-feature-interpretation-via-clustering-based.md", "text": "https://wpnews.pro/news/spice-simple-polysemantic-feature-interpretation-via-clustering-based.txt", "jsonld": "https://wpnews.pro/news/spice-simple-polysemantic-feature-interpretation-via-clustering-based.jsonld"}}