Year: 2025
Type: preprint
Abstract: In-context learning is a remarkable capability of transformers, referring to their ability to adapt to specific tasks based on a short history or context. Previous research has found that task-specific information is locally encoded within models, though their emergence and functionality remain unclear due to opaque pre-training processes. In this work, we investigate the formation of task vectors in a controlled setting, using models trained from scratch on synthetic datasets. Our findings confirm that task vectors naturally emerge under certain conditions, but the tasks may be relatively weakly and/or non-locally encoded within the model. To promote strong task vectors encoded at a prescribed location within the model, we propose an auxiliary training mechanism based on a task vector prompting loss (TVP-loss). This method eliminates the need to search for task-correlated encodings within the trained model and demonstrably improves robustness and generalization.
Source: arXiv (Cornell University)
Authors: Yang, Liu, Lin, Ziqian, Lee, Kangwook, Papailiopoulos, Dimitris, Nowak, Robert
Institutions: University of Wisconsin–Madison, Microsoft (United States)
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Awards: CIF: Small: Advanced Understanding and Applications of Deep Learning, G1660931560, G3587679958, G5121974832, CAREER: Theory and Algorithms for Learning with Frozen Pretrained Models, Collaborative Research: New Perspectives on Deep Learning: Bridging Approximation, Statistical, and Algorithmic Theories, G6671297155, G6706517093, G8876996369
OpenAlex ID: W4406548196