[论文] Designer-RSI: Evolving Procedural Memory from User Traffic for Agentic...

论文概要

研究领域: CV 作者: Hongyang Du, Lan Yan, Christian Flores, Asim Kadav 发布时间: 2026-09-18 arXiv: 2609.22086

中文摘要

专业平面设计是一种长程智能体任务：结构化、可编辑的作品由大量相互依赖的动作逐步生成，但其结果缺乏可靠的程序化判据。我们提出一种持续适应框架：冻结的前沿模型通过 230 多个工具操作专业设计软件，同时一个由自然语言技能构成的外部程序性记忆从经验中积累并精炼可复用的设计流程。记忆通过为反复出现但未覆盖的子任务获取新流程而“横向扩展”，通过对照自身成功与失败的执行修订既有流程而“纵向深化”；一个配对回放门控只接纳能修复失败且不导致已有成功退化的改动。在五轮迭代中，系统处理了 1,406 份真实用户简报和 1,869 条自动评分的轨迹，无需权重更新也无需人工标注，技能库从 76 个文档派生技能增长到 139 个，并将 Claude-Sonnet-4 在 GenEval2 上的执行成功率从 72.7% 提升至 99.3%（生成质量 +11.99 分）；在四个专业设计基准上，对无技能智能体的胜率在 Claude-Sonnet-4 和 Claude-Opus-4.6 上分别为 61.8% 和 67.6%。我们进一步证明两种机制组合有效：在用户流量基准的 200 份留出简报上，仅横向扩展或仅纵向深化的胜率为 49.4% / 48.6%，二者结合达到 58.5%（p = 0.025）。程序性记忆为智能体在噪声大、不可验证反馈下的持续适应提供了一条实用路径。

原文摘要

Professional graphic design is a long-horizon agentic task in which structured, editable artifacts emerge from many interdependent actions, yet outcomes admit no reliable programmatic oracle. We introduce a continual adaptation framework in which a frozen frontier model operates professional design software through more than 230 tools, while an external procedural memory of natural-language skills accumulates and refines reusable design procedures from experience. The memory widens by acquiring procedures for recurring uncovered subtasks and deepens by revising existing procedures against their own successful and failed executions, while a matched replay gate admits only changes that repair failures without regressing observed successes. Five rounds over 1,406 real user briefs and 1,869 au...

#论文 #arXiv #CV #小凯