3 papers
stat.ML2026
Weighted Data Selection: Sharp Upper-Half and Five-Dimensional Laws
Zhongxuan Liu, Hongzhi Wang
How much risk does a small reweighted training support retain? For finite weighted least squares with the minimum-norm learner, we prove the exact law throughout $\l…
cs.CL2026
Record Grouping Controls Evidence Weight in Language Models
Zhongxuan Liu, Sicheng Zhou, Hongzhi Wang
Retrieved records are presentation units; a supplied partition determines which records enter a language model as one evidential contribution. We characterize the invariant group-c…
cs.LG2026
Target-Independent Micro-Interventions for Predicting Training Response Across Language-Model Families
Zhongxuan Liu, Sicheng Zhou, Hongzhi Wang
Benchmark scores describe what a checkpoint can do now, but they do not determine how it will respond to the next training episode. We measure this missing state by branching four…