The Optimal Noise in Noise-Contrastive Learning Is Not What You Think
This work addresses a fundamental problem in self-supervised learning for researchers and practitioners, offering a novel perspective on noise selection that could enhance methods like GANs, though it appears incremental in refining existing frameworks.
The paper challenges the common assumption that the optimal noise distribution in noise-contrastive learning should match the data distribution, showing empirically and theoretically that deviating from this can lead to better statistical estimators with improved asymptotic variance.
Learning a parametric model of a data distribution is a well-known statistical problem that has seen renewed interest as it is brought to scale in deep learning. Framing the problem as a self-supervised task, where data samples are discriminated from noise samples, is at the core of state-of-the-art methods, beginning with Noise-Contrastive Estimation (NCE). Yet, such contrastive learning requires a good noise distribution, which is hard to specify; domain-specific heuristics are therefore widely used. While a comprehensive theory is missing, it is widely assumed that the optimal noise should in practice be made equal to the data, both in distribution and proportion. This setting underlies Generative Adversarial Networks (GANs) in particular. Here, we empirically and theoretically challenge this assumption on the optimal noise. We show that deviating from this assumption can actually lead to better statistical estimators, in terms of asymptotic variance. In particular, the optimal noise distribution is different from the data's and even from a different family.