Invariant Feature Subspace Recovery for Multi-Class Classification
Dec 2, 2022
Speakers
Gargi Balasubramaniam
Speaker · 0 followers
Haoxiang Wang
Speaker · 0 followers
About
Domain generalization aims to learn a model over multiple training environments to generalize to unseen environments. Recently, Wang et al [2022] proposed Invariant-feature Subspace Recovery (ISR), a domain generalization algorithm which uses the means of class-conditional data distributions to provably identify the invariant-feature subspace. However, the original ISR algorithm is conditioned on single class only, without utilizing information from the rest classes. In this work, we consider the setting of multi-class classification, and propose an extension of the ISR algorithm, called ISR-Multiclass. This proposed algorithm can provably recover the invariant-feature subspace with 𝒪(d_spu/k) + 1 environments, where d_spu is the number of spurious features and k is the number of classes. Empirically, we first examine ISR-Multiclass in a synthetic dataset, and demonstrate its superiority over the original ISR in the multi-class setting. Furthermore, we conduct experiments in Multiclass Coloured MNIST, a semi-synthetic dataset with strong spurious correlations, and show that ISR-Multiclass can significantly improve the robustness of neural nets trained by various methods (e.g., ERM and IRM) against spurious correlations.Domain generalization aims to learn a model over multiple training environments to generalize to unseen environments. Recently, Wang et al [2022] proposed Invariant-feature Subspace Recovery (ISR), a domain generalization algorithm which uses the means of class-conditional data distributions to provably identify the invariant-feature subspace. However, the original ISR algorithm is conditioned on single class only, without utilizing information from the rest classes. In this work, we consider th…
Organizer
NeurIPS 2022
Account · 961 followers
Like the format? Trust SlidesLive to capture your next event!
Professional recording and live streaming, delivered globally.
Sharing
Recommended Videos
Presentations on similar topic, category or speaker
Fairness Certificates for Differentially Private Classification
Watch later
Paul Mangold, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Learning Neural Acoustic Fields
Watch later
Andrew Luo, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Gradient dynamics of single-neuron autoencoders on orthogonal data
Watch later
Nikhil Ghosh, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
MetaMask: Revisiting Dimensional Confounder for Self-Supervised Learning
Watch later
Jiangmeng Li, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Understanding Robust Learning Through the Lens of Representation Similarities
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
HAPI: A Large-scale Longitudinal Dataset of Commercial ML API Predictions
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%