From Pixels to Perception: Interpretable Predictions via Instance-wise Grouped Feature Selection
Not on the map. Why?
Abstract (arXiv)
Understanding the decision-making process of machine learning models provides valuable insights into the task, the data, and the reasons behind a model's failures. In this work, we propose a method that performs inherently interpretable predictions through the instance-wise sparsification of input images. To align the sparsification with human perception, we learn the masking in the space of semantically meaningful pixel regions rather than on pixel-level. Additionally, we introduce an explicit way to dynamically determine the required level of sparsity for each instance. We show empirically on semi-synthetic and natural image datasets that our inherently interpretable classifier produces more meaningful, human-understandable predictions than state-of-the-art benchmarks.
Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 46 of its 72 references that are in the map. First posted on arXiv 9 May 2025.
Builds on
46 of 72 references.
References · 46
- 2014Microsoft COCO: Common Objects in Contextcited by 17,608
- 2014ImageNet Large Scale Visual Recognition Challengecited by 8,498
- 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,733
- 2017A Unified Approach to Interpreting Model Predictionscited by 6,555
- 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,070
- 2020Training data-efficient image transformers & distillation through attentioncited by 4,586
- 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesiscited by 4,181
- 2017Axiomatic Attribution for Deep Networkscited by 3,384
- 2019Searching for MobileNetV3cited by 1,994
- 2020Shortcut Learning in Deep Neural Networkscited by 1,969
- 2017Learning Important Features Through Propagating Activation Differencescited by 1,458
- 2017Towards A Rigorous Science of Interpretable Machine Learningcited by 1,416
34 more
- 2018Sanity Checks for Saliency Mapscited by 1,156
- 2016The Concrete Distribution: A Continuous Relaxation of Discrete Random Variablescited by 1,148
- 2020Concept Bottleneck Modelscited by 1,032
- 2014Striving for Simplicity: The All Convolutional Netcited by 990
- 2018RISE: Randomized Input Sampling for Explanation of Black-box Modelscited by 838
- 2018This Looks Like That: Deep Learning for Interpretable Image Recognitioncited by 737
- 2016The Mythos of Model Interpretabilitycited by 721
- 2020Towards Faithfully Interpretable NLP Systems: How should we define and evaluate faithfulness?cited by 599
- 2018A Benchmark for Interpretability Methods in Deep Neural Networkscited by 476
- 2020Noise or Signal: The Role of Image Backgrounds in Object Recognitioncited by 343
- 2018Learning to Explain: An Information-Theoretic Perspective on Model Interpretationcited by 248
- 2020Evaluating Weakly Supervised Object Localization Methods Rightcited by 105
- 2020Stochastic Segmentation Networks: Modelling Spatially Correlated Aleatoric Uncertaintycited by 101
- 2019Benchmarking Attribution Methods with Relative Feature Importancecited by 89
- 2022B-cos Networks: Alignment is All We Need for Interpretabilitycited by 88
- 2015The Bayesian Case Model: A Generative Approach for Case-Based Reasoning and Prototype Classificationcited by 76
- 2021Have We Learned to Explain?: How Interpretability Methods Can Learn to Encode Predictions in their Interpretationscited by 59
- 2023Learning to Maximize Mutual Information for Dynamic Feature Selectioncited by 56
- 2021Shapley values for feature selection: The good, the bad, and the axiomscited by 54
- 2024Stochastic Concept Bottleneck Modelscited by 51
- 2023B-cos Alignment for Inherently Interpretable CNNs and Vision Transformerscited by 31
- 2023Rotating Features for Object Discoverycited by 29
- 2024Beyond Concept Bottleneck Models: How to Make Black Boxes Intervenable?cited by 28
- 2024Interpretable Image Classification with Adaptive Prototype-based Vision Transformerscited by 25
- 2024On the Concept Trustworthiness in Concept Bottleneck Modelscited by 22
- 2017The Doctor Just Won't Accept That!cited by 21
- 2019InfoMask: Masked Variational Latent Representation to Localize Chest Diseasecited by 15
- 2024B-cosification: Transforming Deep Neural Networks to be Inherently Interpretablecited by 12
- 2022Interpretations Steered Network Pruning via Amortized Inferred Saliency Mapscited by 11
- 2023Discriminative Feature Attributions: Bridging Post Hoc Explainability and Inherent Interpretabilitycited by 10
- 2024Comprehensive Attribution: Inherently Explainable Vision Model with Feature Detectorcited by 5
- 2023This Reads Like That: Deep Learning for Interpretable Natural Language Processingcited by 2
- 2024Local Feature Selection without Label or Feature Leakage for Interpretable Machine Learning Predictionscited by 1
- 2024Exploiting Interpretable Capabilities with Concept-Enhanced Diffusion and Prototype Networkscited by 1
Built on by
Citing papers per year: 2026: 2.
2 papers.