Tackling Shortcut Learning in Deep Neural Networks: An Iterative Approach with Interpretable Models
Not on the map. Why?
Abstract (arXiv)
We use concept-based interpretable models to mitigate shortcut learning. Existing methods lack interpretability. Beginning with a Blackbox, we iteratively carve out a mixture of interpretable experts (MoIE) and a residual network. Each expert explains a subset of data using First Order Logic (FOL). While explaining a sample, the FOL from biased BB-derived MoIE detects the shortcut effectively. Finetuning the BB with Metadata Normalization (MDN) eliminates the shortcut. The FOLs from the finetuned-BB-derived MoIE verify the elimination of the shortcut. Our experiments show that MoIE does not hurt the accuracy of the original BB and eliminates shortcuts effectively.
Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 36 of its 44 references that are in the map. First posted on arXiv 20 Feb 2023.
Builds on
36 of 44 references.
References · 36
- 2015Deep Residual Learning for Image Recognitioncited by 52,096
- 2015Distilling the Knowledge in a Neural Networkcited by 10,324
- 2016Densely Connected Convolutional Networkscited by 6,720
- 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,070
- 2014Going Deeper with Convolutionscited by 4,241
- 2020Shortcut Learning in Deep Neural Networkscited by 1,969
- 2019Invariant Risk Minimizationcited by 1,854
- 2017Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)cited by 1,319
- 2019Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalizationcited by 1,247
- 2016Deep CORAL: Correlation Alignment for Deep Domain Adaptationcited by 1,200
- 2020Concept Bottleneck Modelscited by 1,032
- 2020Out-of-Distribution Generalization via Risk Extrapolation (REx)cited by 742
24 more
- 2017Zero-Shot Learning -- A Comprehensive Evaluation of the Good, the Bad and the Uglycited by 538
- 2021Just Train Twice: Improving Group Robustness without Training Group Informationcited by 523
- 2019Learning the Difference that Makes a Difference with Counterfactually-Augmented Datacited by 439
- 2019SelectiveNet: A Deep Neural Network with an Integrated Reject Optioncited by 350
- 2021RadGraph: Extracting Clinical Entities and Relations from Radiology Reportscited by 319
- 2022Post-hoc Concept Bottleneck Modelscited by 293
- 2019Entity, Relation, and Event Extraction with Contextualized Span Representationscited by 262
- 2021Gradient Matching for Domain Generalizationcited by 258
- 2019Adversarial Domain Adaptation with Domain Mixupcited by 250
- 2021Invariance Principle Meets Information Bottleneck for Out-of-Distribution Generalizationcited by 220
- 2022Improving Out-of-Distribution Robustness via Selective Augmentationcited by 202
- 2017Neural-Symbolic Learning and Reasoning: A Survey and Interpretationcited by 196
- 2020A Patient-Centric Dataset of Images and Metadata for Identifying Melanomas Using Clinical Contextcited by 146
- 2022Concept Embedding Models: Beyond the Accuracy-Explainability Trade-Offcited by 61
- 2021Entropy-based Logic Explanations of Neural Networkscited by 51
- 2021Logic Explained Networkscited by 45
- 2020Debiasing Skin Lesion Datasets and Models? Not So Fastcited by 43
- 2021Feature Fusion Vision Transformer for Fine-Grained Visual Categorizationcited by 40
- 2020On Interpretability of Deep Learning based Skin Lesion Classifiers using Concept Activation Vectorscited by 39
- 2022Anatomy-Guided Weakly-Supervised Abnormality Localization in Chest X-rayscited by 21
- 2022Selective Prediction via Training Dynamicscited by 20
- 2020Symbolic Logic meets Machine Learning: A Brief Survey in Infinite Domainscited by 16
- 2021Metadata Normalizationcited by 14
- 2021Patch Shortcuts: Interpretable Proxy Models Efficiently Find Black-Box Vulnerabilitiescited by 4
Built on by
Citing papers per year: 2023: 1; 2024: 0; 2025: 1.
2 papers.