Online Continual Learning with Maximal Interfered Retrieval

TLDR

Continual learning, especially in online single‑pass settings, remains difficult, and existing replay methods that randomly sample memories are suboptimal despite matching state‑of‑the‑art performance on benchmarks. This study aims to investigate controlled sampling of replay memories. The authors propose retrieving the most interfered samples—those whose predictions would be most negatively affected by the upcoming parameter update—and formulate this criterion for both generative and experience replay. The method yields consistent performance gains, markedly reduces forgetting, and is available as an open‑source implementation on GitHub.

Abstract

Continual learning, the setting where a learning agent is faced with a never-ending stream of data, continues to be a great challenge for modern machine learning systems. In particular the online or single-pass through the data setting has gained attention recently as a natural setting that is difficult to tackle. Methods based on replay, either generative or from a stored memory, have been shown to be effective approaches for continual learning, matching or exceeding the state of the art in a number of standard benchmarks. These approaches typically rely on randomly selecting samples from the replay memory or from a generative model, which is suboptimal. In this work, we consider a controlled sampling of memories for replay. We retrieve the samples which are most interfered, i.e. whose prediction will be most negatively impacted by the foreseen parameters update. We show a formulation for this sampling criterion in both the generative replay and the experience replay setting, producing consistent gains in performance and greatly reduced forgetting. We release an implementation of our method at https://github.com/optimass/Maximally_Interfered_Retrieval