NeurIPS 2022

Timezone: »

Poster

MaskTune: Mitigating Spurious Correlations by Forcing to Explore

Saeid Asgari · Aliasghar Khani · Fereshte Khani · Ali Gholami · Linh Tran · Ali Mahdavi Amiri · Ghassan Hamarneh

Thu Dec 01 09:00 AM -- 11:00 AM (PST) @ Hall J #237

in Poster Session 5 »

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. Code for MaskTune is available at https://github.com/aliasgharkhani/Masktune.

Keywords: shortcut learning · spurious correlations · selective classification

[ Paper] [ OpenReview]

Author Information

Saeid Asgari (Autodesk AI)

Aliasghar Khani (Computing Science, Simon Fraser University)

I am a first-year M.Sc. student of CS at SFU, under the supervision of Prof. Ghassan Hamarneh. Recently I was an intern at the AI lab of Autodesk, working under the supervision of Dr. Saeid Asgari. I am mostly interested in fundamental machine learning problems such as shortcut learning and bias. In addition to that, I am inquisitive about self-supervised learning. I am interested in hiking, table tennis, and Iranian musical instruments.

Author Information

Saeid Asgari (Autodesk AI)

Aliasghar Khani (Computing Science, Simon Fraser University)

Fereshte Khani

Ali Gholami (Simon Fraser University)

Linh Tran (UCL / Meta)

Ali Mahdavi Amiri (Simon Fraser University)

Ghassan Hamarneh (Simon Fraser University)

More from the Same Authors