Search

16 results for “ALFWorld”

1. Research · Reinforcement learning

   Hindsight assigns token credit from one verified language-model rollout

   Hindsight-Privileged RLVR: Single-Rollout Token Credit Without Group Sampling

   HP-RLVR uses a trajectory’s realized future and verifier feedback to weight token updates without group sampling, reference answers or a learned critic.

   ALFWorld: 0% zero gradient-norm fraction (as reported)

   15 Oct 2026South China University of Technology, Tsinghua University, The Hong Kong University of Science and Technology

2. Research · Language models

   Branching student trajectories improves on-policy distillation efficiency and accuracy

   Branching On-Policy Distillation

   Chen et al. show that resampling continuations at informative decision points broadens teacher supervision and outperforms independent rollouts with fewer student tokens.

   ALFWorld: 19.1% student-generated token cost saving vs seven independent (as reported)

   12 Oct 2026Tsinghua University, SenseTime Research, Peking University

3. Research · Agents

   Agents improve by distilling hindsight from their own interactions

   Self-Evolving Agents through Privileged Retrospective Distillation

   Retrospective Self-Distillation converts completed trajectories, including failures, into training signals without ground-truth solutions or verifier rewards for policy updates.

   ALFWorld: 76.6% success rate (SR) (as reported)

   21 Oct 2026Meta AI

4. Research · Agents

   Grounding distillation in interaction states improves multi-turn agent performance

   State-OPD: State-Grounded On-Policy Distillation for Multi-Turn Agents

   State-OPD separates executed interaction history from student reasoning, improving teacher supervision and broadening the actions sampled during training.

   ALFWorld: 31.25% success rate (as reported)

   12 Oct 2026Alibaba Group, Tsinghua University, Shanghai Jiao Tong University

5. Research · Language models

   Adaptive distillation triggers teacher takeover when student diverges

   DTDD: Divergence-Triggered Dynamic Distillation for Reliable On-Policy Supervision

   DTDD uses segment-level divergence to switch from on-policy distillation to teacher-generated recovery, improving stability and sample efficiency across reasoning and agentic tasks.

   ALFWorld: 81.32% average success (as reported)

   16 Oct 2026Google, University of Washington, Boston University

6. Research · Reinforcement learning

   Graph-based hindsight improves credit assignment for long-horizon language agents

   GraphHCA: Closed-Form Hindsight Credit Assignment for Long-Horizon LLM Agents

   GraphHCA derives step-level training signals from pooled rollout graphs without a critic, hindsight model, or additional forward pass.

   ALFWorld: 95.7% success rate (as reported)

   29 Sept 2026Beihang University, Zhongguancun Academy, Communication University of China

7. Research · Agents

   Execution graph helps off-the-shelf teachers supervise student agents better

   Graph-Conditioned On-Policy Agent Distillation from Off-the-Shelf Teachers

   Graph-conditioned on-policy distillation scores student actions against indexed teacher executions, improving success on three agent benchmarks.

   ALFWorld: 85.26% success rate (as reported)

   30 Sept 2026Yuanbao Team, Tencent, Tsinghua University, Huazhong University of Science and Technology

8. Research · Agents

   Localized Context Distillation resolves the learning dilemma in interaction experience

   Learning from Interaction Experience via Localized Context Distillation

   The authors propose a method that identifies and revises only critical spans in a student's trajectory, outperforming both off-policy and on-policy context distillation across math, code, and agent tasks.

   ALFWorld: 33.08% task success rate (as reported)

   16 Oct 2026Tencent, The Chinese University of Hong Kong, Zhejiang University

9. Research · Agents

   Diffusion language models enable faster plan generation and repair for long-horizon agents

   Plan-and-Patch: Diffusion Language Models for Agentic Planning

   The Plan-and-Patch framework uses parallel unmasking for planning and infilling for repairs, outperforming autoregressive models in repair success and latency.

   ALFWorld: 28.4% task success rate after reference-assisted repair (as reported)

   10 Oct 2026University of Texas at Austin, Amazon

10. Research · Reinforcement learning

    Shared rollout states improve credit assignment for long-horizon agents

    Cross-Rollout Bellman Closure for Long-Horizon Agentic Reinforcement Learning

    CRBC recursively combines evidence from intersecting rollouts while weighting alternative actions and transitions by their observed frequencies.

    ALFWorld: 96.88% overall success rate (as reported)

    29 Sept 2026Beihang University, Zhongguancun Academy, Communication University of China

11. Research · Agents

    Reverse scoring helps diffusion agents escape repeated failed actions

    Does This Action Still Explain the Task? Reverse Scoring for Diffusion Language Model Agents

    Reflect Reverse reranks candidate actions by how well they explain the task, reducing retry loops without additional training.

    ALFWorld: 16.17% success rate (as reported)

    1 Oct 2026Huawei Noah’s Ark Lab, Technical University of Darmstadt, UCL Centre for AI

12. Research · Agents

    Typed action encoding helps language models learn adaptable agent policies

    Dyad: Extending Large Language Models with Native Typed Decision-Making

    Dyad separates language reasoning from structured action selection, enabling efficient decisions over changing action spaces.

    ALFWorld: 90.9% success rate (as reported)

    30 Sept 2026Zhejiang University, SAP, Central South University

13. Research · Agents

    Agent distillation improves when supervision follows outcomes, not model disagreement

    Teacher-Student Gaps Are Not Enough: Outcome-Guided On-Policy Distillation for Multi-Turn Autonomous Agents

    Outcome-guided weighting identifies which teacher interventions actually help the current student complete multi-turn tasks.

    ALFWorld: 64.2% Overall SR (as reported)

    29 Sept 2026Qwen Large Model Application Team, Alibaba, Peking University, University of Chinese Academy of Sciences

14. Research · Agents

    Explicit belief states help agents stay coherent over long tasks

    Beyond Memory: Harnessing Long-Horizon Agents with Explicit Belief States

    PoS continually revises an agent’s view of the current world, checks it for inconsistencies, and intervenes when progress stalls.

    ALFWorld: 97.01% task success (as reported)

    2 Oct 2026Nankai University, Alibaba Group, Tsinghua University

15. Research · Reinforcement learning

    Online distillation cuts needless verbosity in RL-trained models

    Mitigating the Length-Scaling Tax with Online Distillation

    Length Self-Distillation reduces excess response length on easy queries without sacrificing accuracy on hard ones.

    BrowseComp-Plus: 31.65% Pass@1 (as reported)

    1 Oct 2026ByteDance Seed, Tongji University, Peking University

16. Research · Agents

    Hierarchical supervision allocation improves long-horizon model distillation

    Dense Is Not Enough: Hierarchical Supervision Allocation for Long-Horizon On-Policy Distillation

    The LENS-OPD method outperforms dense token-level supervision by dynamically selecting which student behaviors to correct based on future impact and learnability.

    ALFWorld: 47.9% SR (as reported)

    29 Sept 2026Ant Group, Alibaba International Digital Commerce Group, University of Science and Technology of China