Tag
SkillNEO 提出非精英进化优化方法来改进 LLM agent 技能,通过允许低奖励版本继续迭代并分析版本间差异来保留有效步骤、修复失败,使 GPT-5.4 Mini 在 86 项 SkillsBench 任务上平均奖励从 49.1% 提升至 66.4%,且改进后的技能还能提升更强模型的表现。
提出一个自我进化代理框架(Harness RSI),在保持模型权重固定的情况下,通过反馈驱动的修订与进化优化来持续改进代理的软件脚手架(提示、记忆、工作流和可执行组件),并借助持续 PAC 学习给出泛化与保留的理论保证。
论文提出 CoCoEvolve(CortexCodeEvolve),一种基于"元反思"的进化优化器,通过从完整搜索历史中检索证据并主动探测目标环境来改进生产级 AI 系统的优化,在 DABStep、data-eng-bench 等多个基准上超越现有基线。
VIEWCRAFT 通过联合进化多模态代理的证据界面与读取流程,在八个基准上平均提升 6.1 分任务得分,并以更少的输入 token 实现跨执行器迁移。
HeurEvo introduces an automated plan–code–component co-evolution framework in which AI agents jointly evolve algorithmic structure, executable implementations, and a shared pool of reusable components for hybrid solver-augmented heuristics. Across combinatorial optimization benchmarks and MIPLIB instances, it matches or surpasses state-of-the-art solvers under tight runtime budgets and sets new results on problems such as hexagon packing.
The paper introduces TGL-NSGA-II, a teacher-guided fitness approximation framework for expensive evolutionary optimization in constrained TinyML neural architecture search, achieving improved efficiency and reliability over standard methods.
GraphSkillEvo is an evolutionary optimization framework that represents agent skills as graph-structured artifacts to improve LLM performance, outperforming baselines on multiple benchmarks.
Skill-MAS introduces a method for evolving meta-skills in multi-agent systems to improve orchestration without modifying model weights, achieving transferable performance gains across tasks and LLMs.
Introduces a framework combining flow-based generative editing with evolutionary algorithms to perform optimization in residual space, enabling controllable data editing with non-differentiable objectives. Validated on MorphoMNIST and crystal data.
Introduces an automated prompt optimization framework for LLM game agents that decomposes the observation-to-action pipeline into two agents and iteratively refines prompts via an evolutionary loop guided by environment returns. Evaluated on BabyAI tasks, it significantly improves success rates (e.g., from 0% to 72.5% on PutNext) without updating model weights.
DeepEval introduces an evolutionary optimization method for prompts using genetic algorithms, allowing automatic rewriting based on eval feedback and multi-objective optimization.
This paper proposes a neuroevolution-based fine-tuning method to improve the accuracy of quantized deep learning models, showing that nearest-neighbor rounding alone is suboptimal and that evolutionary mutation of weights can yield better results on architectures like VGG and ResNet.