Data Distributional Properties Drive Emergent In-Context Learning in Transformers

By Stephanie C. Y. Chan, Adam Santoro, Andrew Kyle Lampinen, Jane X. Wang, Aaditya K Singh, Pierre H. Richemond, James L. McClelland, Felix Hill. Neural Information Processing Systems, 2022

389 citations recorded by the source corpus.

Large transformer-based models are able to perform in-context few-shot learning, without being explicitly trained for it. This observation raises the question: what aspects of the training regime lead to this emergent behavior? Here, we show that this behavior is driven by the distributions of the training data itself. In-context learning emerges when the training data exhibits particular distributional properties such as burstiness (items appear in clusters rather than being uniformly distributed over time) and having large numbers of rarely occurring classes. In-context learning also emerges more strongly when item meanings or interpretations are dynamic rather than fixed. These properties are exemplified by natural language, but are also inherent to naturalistic data in a wide range of other domains. They also depart significantly from the uniform, i.i.d. training distributions typically used for standard supervised learning. In our initial experiments, we found that in-context learning traded off against more conventional weight-based learning, and models were unable to achieve both simultaneously. However, our later experiments uncovered that the two modes of learning could co-exist in a single model when it was trained on data following a skewed Zipfian distribution -- another common property of naturalistic data, including language. In further experiments, we found that naturalistic data distributions were only able to elicit in-context learning in transformers, and not in recurrent models. In sum, our findings indicate how the transformer architecture works together with particular properties of the training data to drive the intriguing emergent in-context learning behaviour of large language models, and how future work might encourage both in-context and in-weights learning in domains beyond language.

Subfields: Artificial Intelligence

View Paper: https://www.semanticscholar.org/paper/146e9e1238ff6caf18f0bd936ffcfbe1e65d2afd

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

DataComp: In search of the next generation of multimodal datasets (2023, background): “[19] investigate the effects of dataset distribution on emergent properties of transformers, while Fang et al.”
Are Emergent Abilities of Large Language Models a Mirage? (2023, background): “To emphasize that the sharpness of the metric is responsible for emergent abilities, and to show that sharpness extends to metrics beyond Accuracy, we intentionally define a discontinuous metric that measures a network’s ability to reconstruct a dataset as the average number of test data with…”
Scissorhands: Exploiting the Persistence of Importance Hypothesis for LLM KV Cache Compression at Test Time (2023, background): “Large language models(LLMs), trained on immense amounts of text data, have demonstrated an incredible ability to generate text that is both logically connected and contextually relevant [1–5].”
Towards Revealing the Mystery behind Chain of Thought: a Theoretical Perspective (2023, neutral): “, length extrapolation (Figure 3)) [55, 12].”
Contrastive Distillation Is a Sample-Efficient Self-Supervised Loss Policy for Transfer Learning (2022, neutral): “Chan et al. (2022) show that data distributions similar to those in ﬁrst-person datasets and natural language can induce few-shot learning on Transformer architectures, suggesting that a Transformer-like architecture may have inductive biases for few-shot learning from 1st-person perception data,…”
A Survey of Large Language Models (2023, neutral): “Another study [236] provides an in-depth analysis of the impact of training data distribution.”
Deja Vu: Contextual Sparsity for Efficient LLMs at Inference Time (2023, background): “…number of parameters unleashes impressive performance and emergent in-context-learning abilities—they can perform a task by conditioning on input-output examples, without updating their parameters (Bommasani et al., 2021; Liang et al., 2022; Brown et al., 2020; Min et al., 2022; Chan et al., 2022).”