Multi-label Contrastive Predictive Coding
This addresses a bottleneck in unsupervised representation learning for researchers and practitioners by providing a more accurate mutual information estimator, though it is incremental as it builds on existing contrastive predictive coding methods.
The paper tackled the limitation of mutual information estimators in contrastive predictive coding, which are bounded by log m and can underestimate unless m is large, by introducing a multi-label classification estimator that exceeds this bound and improves estimation. The result showed empirical gains in unsupervised representation learning and outperformed a state-of-the-art knowledge distillation method in 10 out of 13 tasks.
Variational mutual information (MI) estimators are widely used in unsupervised representation learning methods such as contrastive predictive coding (CPC). A lower bound on MI can be obtained from a multi-class classification problem, where a critic attempts to distinguish a positive sample drawn from the underlying joint distribution from $(m-1)$ negative samples drawn from a suitable proposal distribution. Using this approach, MI estimates are bounded above by $\log m$, and could thus severely underestimate unless $m$ is very large. To overcome this limitation, we introduce a novel estimator based on a multi-label classification problem, where the critic needs to jointly identify multiple positive samples at the same time. We show that using the same amount of negative samples, multi-label CPC is able to exceed the $\log m$ bound, while still being a valid lower bound of mutual information. We demonstrate that the proposed approach is able to lead to better mutual information estimation, gain empirical improvements in unsupervised representation learning, and beat a current state-of-the-art knowledge distillation method over 10 out of 13 tasks.