RECORD
MaskTune: Mitigating Spurious Correlations by Forcing to Explore
Machine Learning · Submitted 30 Sep 2022
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim Our empirical results on biased MNIST, CelebA, Waterbirds, and ImagenNet-9L datasets show that MaskTune is effective on tasks that often suffer from the existence of spurious correlations.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
A fundamental challenge of over-parameterized deep learning models is learning meaningful data representations that yield good performance on a downstream task without over-fitting spurious input features. This work proposes MaskTune, a masking strategy that prevents over-reliance on spurious (or a limited number of) features. MaskTune forces the trained model to explore new features during a single epoch finetuning by masking previously discovered features. MaskTune, unlike earlier approaches for mitigating shortcut learning, does not require any supervision, such as annotating spurious features or labels for subgroup samples in a dataset. Our empirical results on biased MNIST, CelebA, Waterbirds, and ImagenNet-9L datasets show that MaskTune is effective on tasks that often suffer from the existence of spurious correlations. Finally, we show that MaskTune outperforms or achieves similar performance to the competing methods when applied to the selective classification (classification with rejection option) task. Code for MaskTune is available at https://github.com/aliasgharkhani/Masktune.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | A fundamental challenge of over-parameterized deep learning models is learning meaningful data representations that yield good performance on a downstream task without over-fitting spurious input features. | background 1.00 | 0.05 | 0.03 | asserted 1.00 |
| s1 | This work proposes MaskTune, a masking strategy that prevents over-reliance on spurious (or a limited number of) features. | contribution 0.95 | 0.85 | 0.18 | asserted 1.00 |
| s2 | MaskTune forces the trained model to explore new features during a single epoch finetuning by masking previously discovered features. | method 0.99 | 0.73 | 0.10 | asserted 1.00 |
| s3 | MaskTune, unlike earlier approaches for mitigating shortcut learning, does not require any supervision, such as annotating spurious features or labels for subgroup samples in a dataset. | contribution 0.80 | 0.48 | 0.07 | asserted 1.00 |
| s4 | Our empirical results on biased MNIST, CelebA, Waterbirds, and ImagenNet-9L datasets show that MaskTune is effective on tasks that often suffer from the existence of spurious correlations. | result 1.00 | 0.97 | 0.43 | asserted 1.00 |
| s5 | Finally, we show that MaskTune outperforms or achieves similar performance to the competing methods when applied to the selective classification (classification with rejection option) task. | result 1.00 | 0.95 | 0.21 | asserted 1.00 |
| s6 | Code for MaskTune is available at https://github.com/aliasgharkhani/Masktune. | contribution 0.73 | 0.41 | 0.02 | asserted 0.99 |
What was examined, and what was not
- arXiv listing abstract v2
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
A fundamental challenge of over-parameterized deep learning models is learning meaningful data representations that yield good performance on a downstream task without over-fitting spurious input features. This work proposes MaskTune, a masking strategy that prevents over-reliance on spurious (or a limited number of) features. MaskTune forces the trained model to explore new features during a single epoch finetuning by masking previously discovered features. MaskTune, unlike earlier approaches for mitigating shortcut learning, does not require any supervision, such as annotating spurious features or labels for subgroup samples in a dataset. Our empirical results on biased MNIST, CelebA, Waterbirds, and ImagenNet-9L datasets show that MaskTune is effective on tasks that often suffer from the existence of spurious correlations. Finally, we show that MaskTune outperforms or achieves similar performance to the competing methods when applied to the selective classification (classification with rejection option) task. Code for MaskTune is available at https://github.com/aliasgharkhani/Masktune.
Human review
How the literature uses it
3 arXiv papers cite this in 4 sentences
Papers citing it · 3 sentences without a confident label
Cites it as background 1 paper
-
One line of work assumes access only to the original biased training set, aiming either to infer spurious attributes implicitly and use them to train a more robust classifier [ref] or emphasize learning a diverse set of features to prevent the model from over-relying on the simplest features [ref] .
Cumulative Meta-Learning from Active Learning Queries for Robustness to Spurious Correlations · Spurious Correlation · 0.98
What it cites · 18 sentences without a confident label
Uses data or values from 2 papers
-
In this experiment, we leverage the CelebA [ref] and the Waterbirds [ref] datasets.
Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization · Classification with Spurious Features · 0.99
-
As a further step, we evaluated on the Background Challenge data [ref] to see if the positive observations from the MNIST experiment apply to a more realistic scenario.
Noise or Signal: The Role of Image Backgrounds in Object Recognition · Classification with Spurious Features · 0.83
Cites as background 5 papers
-
In previous work, a supervised loss function has been employed to reduce the effect of spurious correlations [ref] .
Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization · Introduction · 0.91
-
Other works have attempted to force models to discard context and background (as spurious features) through input morphing using perceptual similarities [ref] or learning casual variables [ref] .
Robust Representation Learning via Perceptual Similarity Metrics · Introduction · 0.81
-
Other works have attempted to force models to discard context and background (as spurious features) through input morphing using perceptual similarities [ref] or learning casual variables [ref] .
Learning Causal Models Online · Introduction · 0.81
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.