On the Out-of-Distribution Generalization of Self-Supervised Learning

TL;DR

This paper investigates the out-of-distribution generalization of self-supervised learning, identifying spurious correlations as a key issue and proposing a causal inference-based sampling strategy to improve OOD robustness.

Contribution

It introduces a novel post-intervention distribution framework and a batch sampling method grounded in causal inference to enhance OOD generalization in SSL.

Findings

Proposed a causal inference framework for SSL OOD analysis

Developed a batch sampling strategy satisfying PID constraints

Validated improved OOD performance through experiments

Abstract

In this paper, we focus on the out-of-distribution (OOD) generalization of self-supervised learning (SSL). By analyzing the mini-batch construction during the SSL training phase, we first give one plausible explanation for SSL having OOD generalization. Then, from the perspective of data generation and causal inference, we analyze and conclude that SSL learns spurious correlations during the training process, which leads to a reduction in OOD generalization. To address this issue, we propose a post-intervention distribution (PID) grounded in the Structural Causal Model. PID offers a scenario where the spurious variable and label variable is mutually independent. Besides, we demonstrate that if each mini-batch during SSL training satisfies PID, the resulting SSL model can achieve optimal worst-case OOD performance. This motivates us to develop a batch sampling strategy that enforces PID constraints through the learning of a latent variable model. Through theoretical analysis, we demonstrate the identifiability of the latent variable model and validate the effectiveness of the proposed sampling strategy. Experiments conducted on various downstream OOD tasks demonstrate the effectiveness of the proposed sampling strategy.