← 研踪 PaperTrailarXiv:2609.22086 · v2
Designer-RSI: Evolving Procedural Memory for Graphic Design Agent
首次提交 2026-09-19 01:59
版本更新 2026-11-19 06:21
雷达观测 2026-11-21 12:18
摘要
Deployed software agents face a continuous stream of user requests while their foundation-model weights remain frozen between releases. Recursive self-improvement (RSI) via external memory offers a practical path to continual adaptation, yet prior methods largely assume programmatic verifiers, cheap retries, or a single global prompt. Creative software breaks these assumptions, as open-ended visual briefs lack ground truth, VLM judges are noisy, and multi-step tool executions are costly and confounded by external assets. We present Designer-RSI, which evolves a modular SKILL.md skill bank from unlabeled design trajectories for an agent spanning over 230 raster, vector, and layout tools. Designer-RSI couples widening (acquiring new skills for recurring capability gaps) with deepening (revising failing skills by contrasting successful and failed runs), and screens updates through a Pareto-style matched replay gate that requires strict gains on sampled briefs without local regressions. Across held-out product briefs, eight external benchmarks, and two backbones, traffic-driven evolution outperforms static documentation skills, lifts execution success up to 99.8%, improves requirement fulfillment, and reaches up to a 67.6% blind win rate over the base agent. Ablations further show that skill acquisition and revision are strongly synergistic: neither suffices alone, and reliable gains require evolving coverage and execution quality together.
作者备注:NeurIPS 2026 CLEA Workshop (Oral)
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-19 06:21 · 最新观测版本
Designer-RSI: Evolving Procedural Memory for Graphic Design Agent
Deployed software agents face a continuous stream of user requests while their foundation-model weights remain frozen between releases. Recursive self-improvement (RSI) via external memory offers a practical path to continual adaptation, yet prior methods largely assume programmatic verifiers, cheap retries, or a single global prompt. Creative software breaks these assumptions, as open-ended visual briefs lack ground truth, VLM judges are noisy, and multi-step tool executions are costly and confounded by external assets. We present Designer-RSI, which evolves a modular SKILL.md skill bank from unlabeled design trajectories for an agent spanning over 230 raster, vector, and layout tools. Designer-RSI couples widening (acquiring new skills for recurring capability gaps) with deepening (revising failing skills by contrasting successful and failed runs), and screens updates through a Pareto-style matched replay gate that requires strict gains on sampled briefs without local regressions. Across held-out product briefs, eight external benchmarks, and two backbones, traffic-driven evolution outperforms static documentation skills, lifts execution success up to 99.8%, improves requirement fulfillment, and reaches up to a 67.6% blind win rate over the base agent. Ablations further show that skill acquisition and revision are strongly synergistic: neither suffices alone, and reliable gains require evolving coverage and execution quality together.
查看此版本解读 →查看此版本原文 ↗