CLA: Latent Alignment for Online Continual Self-Supervised Learning
This addresses the challenge of continual learning in self-supervised settings for applications requiring efficient, real-time adaptation, though it appears incremental as it builds on existing SSL techniques for a specific scenario.
The paper tackles the problem of online continual self-supervised learning, where data arrives in small batches without task boundaries, by introducing Continual Latent Alignment (CLA) to align current and past representations, reducing forgetting and speeding up convergence while outperforming state-of-the-art methods under the same computational budget.
Self-supervised learning (SSL) is able to build latent representations that generalize well to unseen data. However, only a few SSL techniques exist for the online CL setting, where data arrives in small minibatches, the model must comply with a fixed computational budget, and task boundaries are absent. We introduce Continual Latent Alignment (CLA), a novel SSL strategy for Online CL that aligns the representations learned by the current model with past representations to mitigate forgetting. We found that our CLA is able to speed up the convergence of the training process in the online scenario, outperforming state-of-the-art approaches under the same computational budget. Surprisingly, we also discovered that using CLA as a pretraining protocol in the early stages of pretraining leads to a better final performance when compared to a full i.i.d. pretraining.