Abstract
Task-specific fine-tuning can rewrite a language model's answers beyond the training task, complicating updates that must preserve existing behavior. We introduce ATLAS, which turns retained-domain representations into an input-dependent rule for task adaptation. An activation atlas supplies local reference centers and directional filters to a shared low-rank residual. Target supervision learns the residual, while retained geometry shapes its action throughout training and inference. On Qwen3-8B, ATLAS achieves lower mean retained-output Kullback-Leibler (KL) divergence than all seven published baselines at shared coding-performance requirements, with consistent advantages across multiple training seeds. Structural comparisons identify the contributions of retained reference states and directional conditioning, and answer-level analyses show fewer rewritten mathematical answers and more stable commonsense choices. Experiments spanning five backbones and two retained domains further demonstrate coding gains with reduced retained-output movement. With compact storage and modest decoding overhead, ATLAS provides a practical mechanism for acquiring specialized skills while maintaining continuity in existing responses.
Keywords
Subject
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Lin, J., Wei, B., Ding, Y., Liu, S., & Dong, Y. (2026). Learn Here, Move Less Elsewhere: Input-Conditioned Plasticity from Retained-Domain Activation Atlases. https://omanscience.com/en/articles/learn-here-move-less-elsewhere-input-conditioned-plasticity-from-retained-domain-activation-atlases
MLA 9
Lin, Jiangtao, et al. "Learn Here, Move Less Elsewhere: Input-Conditioned Plasticity from Retained-Domain Activation Atlases." https://omanscience.com/en/articles/learn-here-move-less-elsewhere-input-conditioned-plasticity-from-retained-domain-activation-atlases.
Chicago (author–date)
Lin, Jiangtao, Bangyang Wei, Yihang Ding, Siyi Liu, and Yuhan Dong. 2026. "Learn Here, Move Less Elsewhere: Input-Conditioned Plasticity from Retained-Domain Activation Atlases." https://omanscience.com/en/articles/learn-here-move-less-elsewhere-input-conditioned-plasticity-from-retained-domain-activation-atlases.
Harvard
Lin, J., Wei, B., Ding, Y., Liu, S. and Dong, Y. (2026) 'Learn Here, Move Less Elsewhere: Input-Conditioned Plasticity from Retained-Domain Activation Atlases', Available at: https://omanscience.com/en/articles/learn-here-move-less-elsewhere-input-conditioned-plasticity-from-retained-domain-activation-atlases.
Vancouver
Lin J, Wei B, Ding Y, Liu S, Dong Y. Learn Here, Move Less Elsewhere: Input-Conditioned Plasticity from Retained-Domain Activation Atlases. https://omanscience.com/en/articles/learn-here-move-less-elsewhere-input-conditioned-plasticity-from-retained-domain-activation-atlases
IEEE
J. Lin, B. Wei, Y. Ding, S. Liu, and Y. Dong, "Learn Here, Move Less Elsewhere: Input-Conditioned Plasticity from Retained-Domain Activation Atlases," https://omanscience.com/en/articles/learn-here-move-less-elsewhere-input-conditioned-plasticity-from-retained-domain-activation-atlases.