This work takes a fine-grained look at memorization by decomposing the loss trajectory of memorized sequences over training and model parameters, finding that memorization in both cases is characterized by sequence-level gradient alignment.
Abstract
Memorization has been proposed as a mechanism to explain how language models fit the tail of their training distributions, but its training dynamics are not understood well. In this work, we take a fine-grained look at memorization by decomposing the loss trajectory of memorized sequences over training and model parameters. Across the Pythia family, we study memorization of duplicated training sequences (recitation) and rare ones (recollection). We find that memorization in both cases is characterized by sequence-level gradient alignment, though recitation suffers from misalignment with other training influences which causes forgetting, explaining the necessity for higher duplication of these examples. We further show that the lower model layers are the most involved in memorization and forgetting. Predicting memorization, our decomposition improves over a cross-entropy baseline, especially in larger models and early in training. Intervening on a small set of highly influential parameters we are able to ablate memorization in the final model. Together, these findings advance our understanding of how memorization develops during training and offer insights for predicting and intervening on it.
Spaced Repetition Training (SRT) is introduced, a continual learning framework inspired by cognitive science, which schedules sample-rehearsal using the SuperMemo-2 (SM-2) algorithm, and preserves broad benchmark performance that naive continual pre-training and uniform replay substantially degrade.
Together, these results show that composing complementary mechanisms substantially improves long-horizon memorization beyond what any individual mechanism achieves.
Zhe-Yuan Zhang, Alvin Zhang, Daniel Khashabi et al.· 1 citation
We reproduce and stress-test the work of Yu et al. (2023), who characterize how language models (LMs) arbitrate between memorized knowledge and contradictory in-context statements. We replicate their world-capitals experiments on 31 models spanning Pythia, GPT-2, Qwen3, and Ministral families, including base and post-t...
Guilhem Fouilhé, Nicholas Asher, Philippe Muller· Proceedings of the 9th Black...· 0 citations
It is shown that this gating mechanism causes SSMs to first learn an in-weights"memorization"solution, while delaying, or even preventing, convergence to a correct in-context learning solution, and that gating is often beneficial for improving generalization to long sequence lengths.
W. Tong, Aryo Lotfi, Emmanuel Abbé et al.· arXiv.org· 1 citation
This paper first classify memorization- and generalization-heavy models using grokking in modular addition and compare their responses to unlearning, showing that the latter suffer greater retain damage, i.e., a larger performance drop on the retain set.
Hwiyeong Lee, H. Lim, Ingyu Bang et al.· 0 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduNov 17, 2026
By studying the computational foundations of language learning, Associate Professor Jacob Andreas builds intelligent systems that can communicate effectively with humans.