shortcut-learning

Create and manage interactive keyboard shortcut cheatsheets.

Instruction blindness in vision-language-action policies: diagnosis and a low-rank data cure. Paper, model, deconfounded datasets, generator, and the measurement battery.

Figures & code from the paper "Shortcut Learning in Deep Neural Networks" (Nature Machine Intelligence 2020)

[Nature Medicine] The Limits of Fair Medical Imaging AI In Real-World Generalization

How to master your passwords using iCloud Keychain.

Frequency Shortcuts in Neural Networks

TextAdaIN: Paying Attention to Shortcut Learning in Text Recognizers

This is the official code for CoLLAs 2022 paper, "InBiaseD: Inductive Bias Distillation to Improve Generalization and Robustness through Shape-awareness"

Augmentation for CV using frequency shortcuts

UniMod: mitigating shortcut learning in multi-modal medical diagnosis via cross-modality and within-modality alignment (ACM MM 2026)

Study on the effect of masking the ROI in medical images to evaluate potential bias/shortcuts in datasets

GitHub Repository for "Efficient Unsupervised Shortcut Learning Detection and Mitigation in Transformers" presented in ICCV 2025.

My dissertation project proposing novel Hybrid CNN–Transformer A/B models and scalable guidance modules for investigating shortcut-learning mitigation in brain tumour MRI classification when pixel-level masks are unavailable.

Shortcut Wizard is a desktop app that helps professionals and students organize and access application shortcuts effortlessly, boosting productivity with ease.

Referee agents that catch benchmark gaming in clinical multi-agent systems: shortcut cascades, blind metrics, and mutual oversight, measured.

shortcutting: Interactive typing & editing game built to help you learn to quickly use shortcuts while navigating and editing text

A deep learning method using cross-domain regularization to mitigate shortcut learning

AI often 'cheats' to score 100%. This Explainable AI (XAI) project reverse-engineers Neuro-Symbolic models via Causal Abstraction Theory to expose hidden reasoning shortcuts and evaluate architectural fixes for truly trustworthy logic.

Multi-branch CNN for potato leaf disease classification (96.4% on 7 classes, 5 geographic sources) audited with six XAI techniques (Grad-CAM, IG, occlusion, β-routing analysis, k-NN, and counterfactual mask-flip) to distinguish genuine disease recognition from shortcut learning.