---
title: "A-RSI: Adaptive Task Refresh for Zero-Data Multimodal Self-Improvement"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2610.38568v1"
permalink: "https://feed.astralor.com/items/itm_84266c224bf46adbfa9dfba1"
published: "2026-10-28T10:03:42.000Z"
score: 54
channel: "tech"
tags: ["RAG","多模态","A-RSI","Adaptive Task Refresh","Zero-Data Multimodal Self-Improvement","Recursive","Existing","Many"]
---

# A-RSI: Adaptive Task Refresh for Zero-Data Multimodal Self-Improvement

## At a glance

Recursive self-improvement requires models not only to acquire useful training experience, but also to decide when their current experience has ceased to be useful. Existing methods often focus on what experience to gen…

## Key points

- Recursive self-improvement requires models not only to acquire useful training experience, but also to decide when their current experience has ceased to be useful.
- Existing methods often focus on what experience to generate next, while paying less attention to when fresh experience should be introduced.
- Many also construct training targets through self-consistency or majority voting, making supervision depend on the model's own uncertain predictions.

## Full text

Recursive self-improvement requires models not only to acquire useful training experience, but also to decide when their current experience has ceased to be useful. Existing methods often focus on what experience to generate next, while paying less attention to when fresh experience should be introduced. Many also construct training targets through self-consistency or majority voting, making supervision depend on the model's own uncertain predictions. We address these two questions separately: training dynamics determine when to acquire new experience, while the answers to new tasks remain independently checkable. A-RSI monitors reward variation and accuracy in training logs to refresh a task pool when its learning signal declines, without consulting validation or test scores. A static library of model-authored programs produces multimodal questions and derives their reference answers; we specify the task-generation contract, automatically verify chosen properties of the reward and refresh rules, and separately audit the generated answers. The training loop uses no external training examples or human-written answer labels. On the primary eleven-benchmark suite, A-RSI raises the performance average from 57.77 to 61.05 with 800 fresh tasks.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_84266c224bf46adbfa9dfba1