Publication | Conference on Neural Information Processing Systems 2022
MaskTune
Mitigating Spurious Correlations by Forcing to Explore
Activation visualizations of ERM (middle) and MaskTune (right) for Waterbirds samples, in which MaskTune enforces exploring new features. After applying MaskTune, the task-relevant input signals (bird features) are emphasized.
Learning the right features during training is a significant challenge for deep neural networks (DNNs). DNNs might instead pick up spurious features. This work investigates a novel solution to this problem.
Code for MaskTune is available at https://github.com/aliasgharkhani/Masktune.
DownloadAbstract
MaskTune: Mitigating Spurious Correlations by Forcing to Explore
Saeid Asgari, Aliasghar Khani, Fereshte Khani, Ali Gholami, Linh Tran, Ali Mahdavi-Amiri, Ghassan Hamarneh
Conference on Neural Information Processing Systems 2022
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 \method{} outperforms or achieves similar performance to the competing methods when applied to the selective classification (classification with rejection option) task.
Associated Researchers
Saeid Asgari
Former Autodesk
Fereshte Khani
Stanford University
Linh Tran
Autodesk AI Lab
Ghassan Hamarneh
School of Computing Science, Simon Fraser University
Ali Mahdavi-Amiri
School of Computing Science, Simon Fraser University
Related Resources
2024
Evaluating Large Language Models for Material SelectionThis work evaluates the use of LLMs for predicting materials of…
2023
ANPL: Towards Natural Programming with Interactive DecompositionInteractive programming system ensures users can refine generated code…
2024
XLB: A Differentiable Massively Parallel Lattice Boltzmann Library in PythonThis research introduces the XLB library, a scalable Python-based…
2022
Neural Implicit Style-Net: synthesizing shapes in a preferred style exploiting self supervisionWe introduce a novel approach to disentangle style from content in the…
Get in touch
Something pique your interest? Get in touch if you’d like to learn more about Autodesk Research, our projects, people, and potential collaboration opportunities.
Contact us