Emergence of Abstractions: Concept Encoding and Decoding Mechanism for In-Context Learning in Transformers

🤗 Upvotes: 9 | cs.CL, cs.AI, cs.LG Authors: Seungwook Han, Jinyeop Song, Jeff Gore, Pulkit Agrawal Title: Emergence of Abstractions: Concept Encoding and Decoding Mechanism for In-Context Learning in Transformers Arxiv: http://arxiv.org/abs/2412.12276v2 Abstract: Humans distill complex experiences into fundamental abstractions that enable rapid learning and adaptation. Similarly, autoregressive transformers exhibit adaptive learning through in-context learning (ICL), which begs the question of how. In this paper, we propose concept encoding-decoding mechanism to explain ICL by studying how transformers form and use internal abstractions in their representations. On synthetic ICL tasks, we analyze the training dynamics of a small transformer and report the coupled emergence of concept encoding and decoding. As the model learns to encode different latent concepts (e.g., ``Finding the first noun in a sentence.") into distinct, separable representations, it concureently builds conditional decoding algorithms and improve its ICL performance. We validate the existence of this mechanism across pretrained models of varying scales (Gemma-2 2B/9B/27B, Llama-3.1 8B/70B). Further, through mechanistic interventions and controlled finetuning, we demonstrate that the quality of concept encoding is causally related and predictive of ICL performance. Our empirical insights shed light into better understanding the success and failure modes of large language models via their representations.