Paper Detail

Sample-Efficient Learning from Agent Experience

Chenhui Gou, Haoqin Tu, Yunhao Fang, Jianfei Cai, Hamid Rezatofighi

huggingface Score 9.3

Published 2026-07-23 · First seen 2026-07-24

Research Track A · General AI

Abstract

Real-world agent learning is often constrained by costly environment interactions, such as running time-consuming experiments or obtaining human feedback. In-context learning offers a highly sample-efficient way for agents to learn from their own interaction histories, but its gains disappear once that experience is removed from the context. Separately, context distillation provides a mechanism for internalizing contextual information into model weights. However, applying it to agents' interaction histories without sacrificing environment sample efficiency remains underexplored. We term this problem Experience Distillation and develop an implementation that requires no further environment interaction beyond the collected experience. Experiments on 749 curated software-engineering tasks and six text-adventure games show that it retains at least 64.8\% of the gains from in-context learning across both domains, whereas direct supervised fine-tuning on the collected experience recovers only 3.8\%. Compared with classical reinforcement-learning baselines, in-context learning from trial-and-error experience followed by Experience Distillation matches their performance with at least \(9.6\times\) fewer environment samples.

Workflow Status

Review status
pending
Role
unreviewed
Read priority
soon
Vote
Not set.
Saved
no
Collections
Not filed yet.
Next action
Not filled yet.

Reading Brief

No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.

Why It Surfaced

No ranking explanation is available yet.

Tags

No tags.

BibTeX

@misc{gou2026sample,
  title = {Sample-Efficient Learning from Agent Experience},
  author = {Chenhui Gou and Haoqin Tu and Yunhao Fang and Jianfei Cai and Hamid Rezatofighi},
  year = {2026},
  abstract = {Real-world agent learning is often constrained by costly environment interactions, such as running time-consuming experiments or obtaining human feedback. In-context learning offers a highly sample-efficient way for agents to learn from their own interaction histories, but its gains disappear once that experience is removed from the context. Separately, context distillation provides a mechanism for internalizing contextual information into model weights. However, applying it to agents' interacti},
  url = {https://huggingface.co/papers/2607.21051},
  keywords = {huggingface daily},
  eprint = {2607.21051},
  archiveprefix = {arXiv},
}

Metadata

{}