Distilling Model Failures as Directions in Latent Space

06/29/2022
by   Saachi Jain, et al.
0

Existing methods for isolating hard subpopulations and spurious correlations in datasets often require human intervention. This can make these methods labor-intensive and dataset-specific. To address these shortcomings, we present a scalable method for automatically distilling a model's failure modes. Specifically, we harness linear classifiers to identify consistent error patterns, and, in turn, induce a natural representation of these failure modes as directions within the feature space. We demonstrate that this framework allows us to discover and automatically caption challenging subpopulations within the training dataset, and intervene to improve the model's performance on these subpopulations. Code available at https://github.com/MadryLab/failure-directions

READ FULL TEXT

page 9

page 25

page 28

page 29

page 35

page 36

page 37

page 38

research
10/15/2021

Combining Diverse Feature Priors

To improve model generalization, model designers often restrict the feat...
research
07/17/2023

Latent Space Representations of Neural Algorithmic Reasoners

Neural Algorithmic Reasoning (NAR) is a research area focused on designi...
research
09/25/2021

A Principled Approach to Failure Analysis and Model Repairment: Demonstration in Medical Imaging

Machine learning models commonly exhibit unexpected failures post-deploy...
research
10/29/2020

Understanding the Failure Modes of Out-of-Distribution Generalization

Empirical studies suggest that machine learning models often rely on fea...
research
08/13/2022

Self-supervised Matting-specific Portrait Enhancement and Generation

We resolve the ill-posed alpha matting problem from a completely differe...
research
10/29/2021

UDIS: Unsupervised Discovery of Bias in Deep Visual Recognition Models

Deep learning models have been shown to learn spurious correlations from...
research
06/01/2023

Intriguing Properties of Text-guided Diffusion Models

Text-guided diffusion models (TDMs) are widely applied but can fail unex...

Please sign up or login with your details

Forgot password? Click here to reset