Singh, aaditya; Chan, Stephanie CY; Moskovitz, Ted; Grant, erin; Saxe, Andrew; Hill, Felix; (2023) The Transient Nature of Emergent In-context Learning in Transformers. In: Proceedings of the 37th Conference on Neural Information Processing Systems (NeurIPS 2023). NeurIPS (In press).

Preview

Text 8706_the_transient_nature_of_emerge.pdf - Accepted Version Download (4MB) | Preview

Abstract

Transformer neural networks can exhibit a surprising capacity for in-context learning (ICL), despite not being explicitly trained for it. Prior work has provided a deeper understanding of how ICL emerges in transformers, e.g., through the lens of mechanistic interpretability, Bayesian inference, or by examining the distributional properties of training data. However, in each of these cases, ICL is treated largely as a persistent phenomenon; namely, once ICL emerges, it is assumed to persist asymptotically. Here, we show that the emergence of ICL during transformer training is, in fact, often transient. We train transformers on synthetic data designed so that both ICL or in-weights learning (IWL) strategies can lead to correct predictions. We find that ICL first emerges, then disappears and gives way to IWL, all while the training loss decreases, indicating an asymptotic preference for IWL. The transient nature of ICL is observed in transformers across a range of model sizes and datasets, raising the question of how much to “overtrain” transformers when seeking compact, cheaper-to-run models. We find that L2 regularization may offer a path to more persistent ICL that removes the need for early stopping based on ICL-style validation tasks.

Download activity - last month

Export as XMLJSONCSV

Download activity - last 12 months

Export as XMLJSONCSV

Downloads by country - last 12 months

Export as XMLJSONCSV

Archive Staff Only

View Item