Tag
This paper introduces a method for monitoring the reasoning process of Large Reasoning Models by analyzing probe trajectories—the evolution of a concept's probability across generated tokens. The approach uses temporal and signal-processing features from hidden representations to better predict future model behavior, achieving up to 95% AUROC with max-pooling.
Proposes Selective Alignment Knowledge Distillation (SeAl-KD) for Spiking Neural Networks, which selectively aligns class-level and temporal knowledge by equalizing competing logits at erroneous timesteps and reweighting temporal alignment based on confidence and inter-timestep similarity, achieving consistent improvements over existing distillation methods on static and neuromorphic datasets.