…This motivates treating teacher exposure not as a fixed hyperparameter but as a learnable training-time control variable. We therefore propose Adaptive Teacher Exposure for Self-Distillation (ATESD). ATESD models the reveal…
…Existing scalar, score-token, and pairwise reward models over-compress uncertainty and fine-grained score differences, while reasoning-based generative rewards provide stronger judgments but are costly to deploy and difficult to…
…Reinforcement Learning Unlocks Parametric Knowledge in LLMs (2026) Step-wise Rubric Rewards for LLM Reasoning (2026) Confidence-Orchestrated Self-Evolution against Uncertain LLM Feedback (2026) Think Through Uncertainty: Improving Long-Form Generation…
…Exploration on Adaptively Reformulated Instances Enables Learning from Hard Reasoning Problems (2026) What If Consensus Lies? Selective-Complementary Reinforcement Learning at Test Time (2026) Think Through Uncertainty: Improving Long-Form Generation Factuality…
…The editor sees its target failure modes during training instead of chasing a generic uncertainty signal at decode time. the shared-prefix KV caching paired with Alternating Step Decode in the reflective…
…We present TIDE, the first framework for cross-architecture dLLM distillation , comprising three modular components: (1) TIDAL , which jointly modulates distillation strength across training progress and diffusion timestep to account for the…
…with TAURA (Turn-aware Adaptive Uncertainty Rescaled Advantage), which leverages turn-level entropy to steer credit assignment toward pivotal discovery turns. Crucially, OmniAgent exhibits positive test-time scaling, where performance improves as…
To show you the most relevant results, we’ve omitted some entries very similar to those already shown. Repeat the search with the omitted results included.