2 papers
cs.AI2026
From Soft Targets to Reward Signals: How Assignment and Reward Objectives Interact
Jiangtao Lin, Bangyang Wei, Siyi Liu +2
Soft preference targets specify supervision strength, and reward objectives convert that strength into learned reward signals. A central design question remains: how does assigning…
cs.LG2026
Learn Here, Move Less Elsewhere: Input-Conditioned Plasticity from Retained-Domain Activation Atlases
Jiangtao Lin, Bangyang Wei, Yihang Ding +2
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 turn…