mlsanityData to 7 Oct 2026

Understanding Inhibition Through Maximally Tense Images

arXiv 2406.05598 ↗First posted 8 Jun 2024cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

We address the functional role of 'feature inhibition' in vision models; that is, what are the mechanisms by which a neural network ensures images do not express a given feature? We observe that standard interpretability tools in the literature are not immediately suited to the inhibitory case, given the asymmetry introduced by the ReLU activation function. Given this, we propose inhibition be understood through a study of 'maximally tense images' (MTIs), i.e. those images that excite and inhibit a given feature simultaneously. We show how MTIs can be studied with two novel visualization techniques; +/- attribution inversions, which split single images into excitatory and inhibitory components, and the attribution atlas, which provides a global visualization of the various ways images can excite/inhibit a feature. Finally, we explore the difficulties introduced by superposition, as such interfering features induce the same attribution motif as MTIs.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 35 of its 63 references that are in the map. First posted on arXiv 8 Jun 2024.

Builds on

35 of 63 references.

References · 35

  1. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,160
  2. 2017A Unified Approach to Interpreting Model Predictionscited by 6,488
  3. 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,040
  4. 2014Going Deeper with Convolutionscited by 4,231
  5. 2017Axiomatic Attribution for Deep Networkscited by 3,350
  6. 2018UMAP: Uniform Manifold Approximation and Projection for Dimension Reductioncited by 2,841
  7. 2013Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Mapscited by 2,348
  8. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,359
  9. 2017Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)cited by 1,309
  10. 2017SmoothGrad: removing noise by adding noisecited by 1,097
  11. 2022Toy Models of Superpositioncited by 928
  12. 2018Robustness May Be at Odds with Accuracycited by 873
23 more
  1. 2018RISE: Randomized Input Sampling for Explanation of Black-box Modelscited by 825
  2. 2018SNIP: Single-shot Network Pruning based on Connection Sensitivitycited by 802
  3. 2014Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Imagescited by 774
  4. 2017Interpretable Explanations of Black Boxes by Meaningful Perturbationcited by 611
  5. 2014Understanding Deep Image Representations by Inverting Themcited by 399
  6. 2016Not Just a Black Box: Learning Important Features Through Propagating Activation Differencescited by 295
  7. 2013k-Sparse Autoencoderscited by 209
  8. 2016Synthesizing the preferred inputs for neurons in neural networks via deep generator networkscited by 181
  9. 2022CRAFT: Concept Recursive Activation FacTorization for Explainabilitycited by 177
  10. 2017Visualizing Deep Neural Network Decisions: Prediction Difference Analysiscited by 166
  11. 2020Invertible Concept-based Explanations for CNN Models with Non-negative Concept Activation Vectorscited by 119
  12. 2023A Holistic Approach to Unifying Automatic Concept Extraction and Concept Importance Estimationcited by 108
  13. 2016Multifaceted Feature Visualization: Uncovering the Different Types of Features Learned By Each Neuron in Deep Neural Networkscited by 106
  14. 2021Look at the Variance! Efficient Black-box Explanations with Sobol-based Sensitivity Analysiscited by 65
  15. 2019How Can We Be So Dense? The Benefits of Using Highly Sparse Representationscited by 45
  16. 2022Don't Lie to Me! Robust and Efficient Explainability with Verified Perturbation Analysiscited by 38
  17. 2022Making Sense of Dependence: Efficient Black-box Explanations Using Dependence Measurecited by 33
  18. 2021How Well do Feature Visualizations Support Causal Understanding of CNN Activations?cited by 28
  19. 2023Unlocking Feature Visualization for Deeper Networks with MAgnitude Constrained Optimizationcited by 26
  20. 2015Understanding Intra-Class Knowledge Inside CNNcited by 22
  21. 2023Identifying Interpretable Visual Features in Artificial and Biological Neural Systemscited by 9
  22. 2024Feature Accentuation: Revealing 'What' Features Respond to in Natural Imagescited by 5
  23. 2022Why do networks have inhibitory/negative connections?cited by 3

Built on by

Not cited yet.