mlsanityData to 14 Oct 2026

PURE: Turning Polysemantic Neurons Into Pure Features by Identifying Relevant Circuits

arXiv 2404.06453 ↗First posted 9 Apr 2024cs.CV cs.AI cs.LGcited by 29

On the map: Mechanistic interpretability → Circuits and mechanistic analysis

Abstract (arXiv)

The field of mechanistic interpretability aims to study the role of individual neurons in Deep Neural Networks. Single neurons, however, have the capability to act polysemantically and encode for multiple (unrelated) features, which renders their interpretation difficult. We present a method for disentangling polysemanticity of any Deep Neural Network by decomposing a polysemantic neuron into multiple monosemantic "virtual" neurons. This is achieved by identifying the relevant sub-graph ("circuit") for each "pure" feature. We demonstrate how our approach allows us to find and disentangle various polysemantic units of ResNet models trained on ImageNet. While evaluating feature visualizations using CLIP, our method effectively disentangles representations, improving upon methods based on neuron activations. Our code is available at https://github.com/maxdreyer/PURE.

Key facts. Cited by 29 papers in the map (data through 14 Oct 2026). Placed in Circuits and mechanistic analysis, one of 13 directions in Mechanistic interpretability. Builds on 15 of its 30 references that are in the map. First posted on arXiv 9 Apr 2024.

Builds on

15 of 30 references.

Closest ancestors · 2

Same direction, newest first

  1. 2023Towards Automated Circuit Discovery for Mechanistic Interpretability28 Apr
  2. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small1 Nov

Other references · 13

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  2. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,974
  3. 2017Learning Important Features Through Propagating Activation Differencescited by 1,458
  4. 2017Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)cited by 1,319
  5. 2021ResNet strikes back: An improved training procedure in timmcited by 368
  6. 2023How does GPT-2 compute greater-than?: Interpreting mathematical abilities in a pre-trained language modelcited by 202
  7. 2023A Holistic Approach to Unifying Automatic Concept Extraction and Concept Importance Estimationcited by 109
  8. 2016Multifaceted Feature Visualization: Uncovering the Different Types of Features Learned By Each Neuron in Deep Neural Networkscited by 106
  9. 2023Multi-dimensional concept discovery (MCD): A unifying framework with completeness guaranteescited by 50
  10. 2023Identifying Interpretable Subspaces in Image Representationscited by 32
  11. 2023Labeling Neural Representations with Inverse Recognitioncited by 32
  12. 2023Understanding the (Extra-)Ordinary: Validating Deep Model Decisions with Prototypical Concept-based Explanationscited by 22
1 more
  1. 2022DORA: Exploring Outlier Representations in Deep Neural Networkscited by 14

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 2; 2025: 18; 2026: 9.

29 papers.

Sort:

By area:

  1. 2025Everything, Everywhere, All at Once: Is Mechanistic Interpretability Identifiable?cited by 34
  2. 2025Mechanistic understanding and validation of large AI models with SemanticLenscited by 29
  3. 2024Style-Specific Neurons for Steering LLMs in Text Style Transfercited by 24
  4. 2025When Truthful Representations Flip Under Deceptive Instructions?cited by 10
  5. 2025ConSim: Measuring Concept-Based Explanations' Effectiveness with Automated Simulatabilitycited by 9
  6. 2025CoE: Chain-of-Explanation via Automatic Visual Concept Circuit Description and Polysemanticity Quantificationcited by 9
  7. 2025Disentangling Polysemantic Channels in Convolutional Neural Networkscited by 6
  8. 2026Certified Circuits: Stability Guarantees for Mechanistic Circuitscited by 5
  9. 2025Signal in the Noise: Polysemantic Interference Transfers and Predicts Cross-Model Influencecited by 4
  10. 2026Spectral Superposition: A Theory of Feature Geometrycited by 4
  11. 2025Expand Neurons, Not Parameterscited by 3
  12. 2025Ensuring Medical AI Safety: Interpretability-Driven Detection and Mitigation of Spurious Model Behavior and Associated Datacited by 2
17 more
  1. 2025SAFR: Neuron Redistribution for Interpretabilitycited by 2
  2. 2025See What I Mean? CUE: A Cognitive Model of Understanding Explanationscited by 2
  3. 2025Mammo-SAE: Interpreting Breast Cancer Concept Learning with Sparse Autoencoderscited by 2
  4. 2025How Robust is Model Editing after Fine-Tuning? An Empirical Study on Text-to-Image Diffusion Modelscited by 1
  5. 2025Disentangling Polysemantic Neurons with a Null-Calibrated Polysemanticity Index and Causal Patch Interventionscited by 1
  6. 2025FaCT: Faithful Concept Traces for Explaining Neural Network Decisionscited by 1
  7. 2026Seeing Through Circuits: Faithful Mechanistic Interpretability for Vision Transformerscited by 1
  8. 2026Layerwise Convergence Fingerprints for Runtime Misbehavior Detection in Large Language Modelscited by 1
  9. 2024Identifying Sub-networks in Neural Networks via Functionally Similar Representationscited by 0
  10. 2025Attack logics, not outputs: Towards efficient robustification of deep neural networks by falsifying concept-based propertiescited by 0
  11. 2025Open Vocabulary Compositional Explanations for Neuron Alignmentcited by 0
  12. 2025Guaranteed Optimal Compositional Explanations for Neuronscited by 0
  13. 2026Shortcut Mitigation via Spurious-Positive Samplescited by 0
  14. 2026Capability neq Interpretability: Human Interpretability of Vision Foundation Modelscited by 0
  15. 2026Interpretability Without Tradeoffs: Disentangling Polysemanticity At Equal Predictive Performancecited by 0
  16. 2026Sparsity, Superposition, and Forgetting: A Mechanistic Study of Representation Retention in Continual Learningcited by 0
  17. 2026SPICE: Simple Polysemantic Feature Interpretation via Clustering-based Explanationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2023Disentangling Neuron Representations with Concept Vectors
  2. 2022Interpreting Neural Networks through the Polytope Lens
  3. 2024Wasserstein Distances, Neuronal Entanglement, and Sparsity
  4. 2023Identifying Interpretable Visual Features in Artificial and Biological Neural Systems
  5. 2025Faithful and Stable Neuron Explanations for Trustworthy Mechanistic Interpretability
  6. 2025Towards Combinatorial Interpretability of Neural Computation
  7. 2026Mechanistic Interpretability for Neural Networks: Circuits, Sparse Features and Symbolic Reasoning
  8. 2022Toy Models of Superposition