Julianna Piskorz, Antonin Berthon +1Affiliation: University of Cambridge, Affiliation: Cambridge, UK
Investigates how rollout policy, token-level KL direction, and learning rate each affect LLM distillation across Llama3 and Qwen2.5 on reasoning tasks; finds KL direction and learning rate dominate outcomes while rollout policy has a modest effect.
AI Agent Papers2026
Records structural priors with skill-specific policies so a runtime agent can select and compose the version of each skill best suited to new states, improving out-of-distribution and compositional generalization for robot manipulation from few demonstrations.
Normalizes each domain's teacher feedback spread during multi-teacher on-policy distillation so that no domain (e.g., instruction following) overwhelms others, improving student recovery of specialist skills and raising average scores across benchmarks.
Trains a single unified multimodal model with reinforcement learning to perform end-to-end self-reflection and iterative image repair — jointly learning the diagnostic (textual) reflection and the flow-based image revisions so credit propagates across rounds without an external verifier.
AI Video Papers2026
Songlin Yang, Xiaotong Zhao +6MMLab@HKUST, The Hong Kong University of Science and Technology, Tencent Video +1
Proposes a forward-process RL method that dynamically localizes gradient updates and adapts multi-reward coordination for joint audio–video diffusion models. Uses bidirectional cross-attention responses for token/layer routing and preference-preserving reweighting to balance competing objectives, improving modality quality, alignment, and synchronization.
Extrapolates RL-induced representation residuals into student hidden states during on-policy distillation: at each layer and token, it regresses the student beyond the teacher along the teacher’s RL-induced direction to improve stability and empirical performance.
Tong Zheng, Skylar Zhai +8University of Chinese Academy of Sciences, University of Minnesota Twin Cities +6
Calibrates multi-reward reinforcement learning by adaptively upweighting infrequently active rewards per rollout batch, so sparse objectives provide stronger signals when they matter. Proposes an inverse-square-root density correction and shows faster learning on tool-calling and math-reasoning tasks.
Generates quantized point-and-box visual primitives for precise grounding across detection, referring, OCR, GUI and layout tasks; uses a shared coordinate vocabulary and staged pretraining to improve downstream robotic manipulation and autonomous driving when used as a visual backbone.
Demonstrates that additive weighted aggregation in rubric-based reinforcement learning enables reward hacking—models improve rubric coverage but produce worse answers—and proposes ProRubric, an offline regrouping that aggregates checklist criteria into conjunctive protocol-level dimensions to raise appropriateness without losing coverage.
Adapts LLM agents online by self-distilling verified execution trajectories into persistent LoRA weights during deployment to improve success and efficiency on long‑horizon tasks. Uses a frozen stable copy as a privileged teacher to predict hindsight next‑token distributions and filters invalid-action turns so experience consolidates without external solutions or memory retrieval.
AI Agent Papers2026
Verifies and preserves trajectory-derived skill edits for LLM agents by pairing each proposed edit with replayable execution evidence and re-executing the relevant trajectory segments. Introduces Replayable Evidence Cards, a replay-based verification gate, and a Provisional Edit Ledger to retain locally supported edits across epochs for continual skill evolution.
AI Video Papers2026
Generates 5-second text- or image-conditioned videos with synchronized 44 kHz audio (including lip-sync) and built-in super-resolution to 1920×1080; available in Lite (3B) and Pro (29B) variants with code and checkpoints released under an MIT license.