MadryLab / failure-directions
Distilling Model Failures as Directions in Latent Space
☆45Updated last year
Related projects: ⓘ
- A modern look at the relationship between sharpness and generalization [ICML 2023]☆42Updated last year
- Code for the ICLR 2022 paper. Salient Imagenet: How to discover spurious features in deep learning?☆35Updated 2 years ago
- Simple data balancing baselines for worst-group-accuracy benchmarks.☆39Updated 10 months ago
- ☆31Updated 7 months ago
- ☆55Updated 4 years ago
- ☆12Updated 6 months ago
- Code for the paper "The Journey, Not the Destination: How Data Guides Diffusion Models"☆19Updated 9 months ago
- [ICLR'22] Self-supervised learning optimally robust representations for domain shift.☆23Updated 2 years ago
- ☆40Updated last year
- ☆102Updated 11 months ago
- Spurious Features Everywhere - Large-Scale Detection of Harmful Spurious Features in ImageNet☆29Updated last year
- ModelDiff: A Framework for Comparing Learning Algorithms☆52Updated last year
- Towards Understanding Sharpness-Aware Minimization [ICML 2022]☆34Updated 2 years ago
- An Investigation of Why Overparameterization Exacerbates Spurious Correlations☆30Updated 4 years ago
- ☆34Updated 3 years ago
- Source code of "What can linearized neural networks actually say about generalization?☆17Updated 2 years ago
- Do input gradients highlight discriminative features? [NeurIPS 2021] (https://arxiv.org/abs/2102.12781)☆13Updated last year
- ☆21Updated last year
- Invariant-feature Subspace Recovery (ISR)☆22Updated last year
- ☆37Updated 3 years ago
- ☆23Updated last year
- Training vision models with full-batch gradient descent and regularization☆37Updated last year
- ☆17Updated last year
- Code for the paper "Evading Black-box Classifiers Without Breaking Eggs" [SaTML 2024]☆19Updated 5 months ago
- Code for "Just Train Twice: Improving Group Robustness without Training Group Information"☆66Updated 4 months ago
- Code for Environment Inference for Invariant Learning (ICML 2021 Paper)☆48Updated 3 years ago
- ☆39Updated last year
- Dataset Interfaces: Diagnosing Model Failures Using Controllable Counterfactual Generation☆43Updated last year
- [NeurIPS'22] Official Repository for Characterizing Datapoints via Second-Split Forgetting☆14Updated last year
- ☆16Updated last year