FEATURE RECONSTRUCTION FROM OUTPUTS CAN MITIGATE SIMPLICITY BIAS IN NEURAL NETWORKS

Addepalli, S and Nasery, A and Netrapalli, P and Babu, VR and Jain, P (2023) FEATURE RECONSTRUCTION FROM OUTPUTS CAN MITIGATE SIMPLICITY BIAS IN NEURAL NETWORKS. In: 11th International Conference on Learning Representations, ICLR 2023, 1 May 2023through 5 May 2023, Kigali.

PDF int_con_lea_rep_2023.pdf.crdownload - Published Version Restricted to Registered users only Download (20MB) | Request a copy

Abstract

Deep Neural Networks are known to be brittle to even minor distribution shifts compared to the training distribution. While one line of work has demonstrated that Simplicity Bias (SB) of DNNs - bias towards learning only the simplest features - is a key reason for this brittleness, another recent line of work has surprisingly found that diverse/complex features are indeed learned by the backbone, and their brittleness is due to the linear classification head relying primarily on the simplest features. To bridge the gap between these two lines of work, we first hypothesize and verify that while SB may not altogether preclude learning complex features, it amplifies simpler features over complex ones. Namely, simple features are replicated several times in the learned representations while complex features might not be replicated. This phenomenon, we term Feature Replication Hypothesis, coupled with the Implicit Bias of SGD to converge to maximum margin solutions in the feature space, leads the models to rely mostly on the simple features for classification. To mitigate this bias, we propose Feature Reconstruction Regularizer (FRR) to ensure that the learned features can be reconstructed back from the logits. The use of FRR in linear layer training (FRR-L) encourages the use of more diverse features for classification. We further propose to finetune the full network by freezing the weights of the linear layer trained using FRR-L, to refine the learned features, making them more suitable for classification. Using this simple solution, we demonstrate up to 15 gains in OOD accuracy on the recently introduced semi-synthetic datasets with extreme distribution shifts. Moreover, we demonstrate noteworthy gains over existing SOTA methods on the standard OOD benchmark DomainBed as well. © 2023 11th International Conference on Learning Representations, ICLR 2023. All rights reserved.

Item Type:	Conference Paper
Publication:	11th International Conference on Learning Representations, ICLR 2023
Publisher:	International Conference on Learning Representations, ICLR
Additional Information:	The copyright for this article belongs to publisher.
Keywords:	Brittleness; Complex networks; Deep neural networks; Fracture mechanics; Plasticity, Diverse features; Feature reconstruction; Feature space; Linear classification; Maximum margin; Neural-networks; Regularizer; Simple++; Training features; Two-line, Classification (of information)
Department/Centre:	Division of Interdisciplinary Sciences > Computational and Data Sciences
Date Deposited:	12 Oct 2024 09:52
Last Modified:	12 Oct 2024 09:52
URI:	http://eprints.iisc.ac.in/id/eprint/86430

Actions (login required)

View Item