Newshot
Policy & Society 中文 Event #01M7PN37V7ES4YTCM78RKRD7QN

Streaming On-Policy Distillation: Decoupling Three Clocks for Stage-Wise Supervision Consumption

Event overview(model summary, not the source's own text)

Streaming On-Policy Distillation: Decoupling Three Clocks for Stage-Wise Supervision Consumption — 3 reports from 2 sources so far.

How this story unfolded

Left: what it concluded — the key points the model distilled from the reports below, each one listing the reports it rests on. Right: what happened — ordered by time, each with a short summary so you can follow the thread without clicking through. The numbered links work both ways: tap a source under a point to jump to that report, tap the cited-by chip on a report to jump back to the point.

Key points0 items Model summary · ordered by when the evidence was published

  1. Only a handful of reports so far — the model has not distilled any key points yet.

ReportsBy publish time · newer further down

  1. November 18, 2026 · Wed
  2. Start

    Streaming On-Policy Distillation: Decoupling Three Clocks for Stage-Wise Supervision Consumption

    Original summary arXiv:2611.22298v1 Announce Type: new Abstract: In on-policy distillation (OPD), response completion, supervision delivery, and parameter updates need not share the same granularit

    arXiv cs.AI Original ↗
  3. November 19, 2026 · Thu
  4. +1d 0h

    Which Tokens, in What Order, and How Fast? A Theoretical Analysis of On-Policy Distillation Objectives

    Original summary arXiv:2611.23397v1 Announce Type: new Abstract: On-policy distillation (OPD) trains a student language model to match a teacher model based on its own generations. When using OPD,

    arXiv cs.LG Original ↗ Headline only
  5. +1d 0h

    Communicative On-Policy Distillation: Distilling Black-Box Teacher Capabilities through Communication for Long-Horizon Tasks

    Original summary arXiv:2611.23673v1 Announce Type: new Abstract: On-policy distillation (OPD) is a powerful paradigm for improving language-model agents on long-horizon tasks, yet its conventional

    arXiv cs.LG Original ↗ Headline only

Related reading

On the Utility of Pre-Trained Visual Representations for Policy Learning of Contact-Rich Insertion Tasks
arXiv:2611.17631v1 Announce Type: new Abstract: Visuomotor imitation learning has become a predominant paradigm for real
arXiv cs.RO 11/16/2026 Full text here
62pts
MINT: Meeting-time Indicators for Truncation in Multi-Step Off-Policy RL
arXiv:2611.24413v1 Announce Type: new Abstract: Many multi-step off-policy temporal-difference methods handle policy mis
arXiv cs.LG 11/19/2026 Full text here
77pts
ExplicitIR: Explicit Spatial Policy Fields for All-in-One Image Restoration
arXiv:2611.22222v1 Announce Type: new Abstract: All-in-one image restoration aims to recover clean images blindly from i
arXiv cs.CV 11/18/2026 Full text here
77pts
Replay, Don't Resample: Off-Policy Consolidation of Long-Horizon RL Experts
arXiv:2611.24059v1 Announce Type: new Abstract: Frontier post-training follows a common recipe: train one expert per dom
arXiv cs.LG 11/19/2026 Full text here
77pts
DexFormer 2.0: Towards a Cross-Embodiment Foundation Policy for Dexterous Grasping
arXiv:2611.20273v2 Announce Type: new Abstract: Dexterous grasping policies are usually trained for one arm paired with
arXiv cs.RO 11/18/2026 Full text here
72pts
Value and Policy Calibration from Execution Feedback for Socially Compatible Driving
arXiv:2611.19969v1 Announce Type: new Abstract: Autonomous vehicle (AV) decisions at intersections affect the delay, bra
arXiv cs.RO 11/18/2026 Full text here
72pts