Paper Registry
Reading: 2 of 5 · After browsing this, you’ll have a categorized map of the field — 50+ papers across 9 categories, with priority ratings to guide what to read first.
Quick Navigation
9 categories: Evolutionary Foundations · Self-Evolving Agents · Workflow Evolution · Quality-Diversity · Multi-Agent Collaboration · Prompt Optimization · Evaluation · Failure Modes · Cross-Domain. Look for !!! = must read.
All identified papers, categorized and prioritized. As of: 2026-03-19.
Legend:
- !!! = Must read (directly relevant)
- !! = Important (relevant methodology or insight)
- ! = Background (useful context)
- CN = Chinese authors / institutions
Paper-Registry
Lesehinweis: 2 von 5 · Nach dem Durchstöbern hast du eine kategorisierte Übersicht des Feldes — 50+ Papers in 9 Kategorien, mit Prioritätsbewertungen als Leseleitfaden.
Schnellnavigation
9 Kategorien: Evolutionäre Grundlagen · Selbst-evolvierende Agents · Workflow-Evolution · Quality-Diversity · Multi-Agent-Kollaboration · Prompt-Optimierung · Evaluation · Fehlermodi · Cross-Domain. Achte auf !!! = Pflichtlektüre.
Alle identifizierten Papers, kategorisiert und priorisiert. Stand: 2026-03-19.
Legende:
- !!! = Muss gelesen werden (direkt relevant)
- !! = Wichtig (relevante Methodik oder Erkenntnis)
- ! = Hintergrund (nützlicher Kontext)
- CN = Chinesische Autor:innen / Institutionen
Category 1: Evolutionary Foundations
The mathematical basis — Nowak's equations, quasispecies theory, evolutionary dynamics.
| Prio | Paper | Authors | Venue | Year | ID |
|---|---|---|---|---|---|
| !!! | Evolutionary Dynamics: Exploring the Equations of Life | Nowak | Harvard UP (Book) | 2006 | — |
| !!! | Prevolutionary dynamics and the origin of evolution | Nowak, Ohtsuki | PNAS | 2008 | 10.1073/pnas.0806714105 |
| !! | Originator dynamics | Manapat, Ohtsuki, Bürger, Nowak | J Theor Biol | 2009 | — |
| !! | Evolutionary Systems Thinking — From Equilibrium to Adaptive Dynamics | Adler | arXiv | 2026 | 2602.15957 |
Kategorie 1: Evolutionäre Grundlagen
Die mathematische Basis — Nowaks Gleichungen, Quasispezies-Theorie, Evolutionsdynamik.
| Prio | Paper | Autoren | Venue | Jahr | ID |
|---|---|---|---|---|---|
| !!! | Evolutionary Dynamics: Exploring the Equations of Life | Nowak | Harvard UP (Buch) | 2006 | — |
| !!! | Prevolutionary dynamics and the origin of evolution | Nowak, Ohtsuki | PNAS | 2008 | 10.1073/pnas.0806714105 |
| !! | Originator dynamics | Manapat, Ohtsuki, Bürger, Nowak | J Theor Biol | 2009 | — |
| !! | Evolutionary Systems Thinking — From Equilibrium to Adaptive Dynamics | Adler | arXiv | 2026 | 2602.15957 |
Category 2: Quality-Diversity / Evolvability
Algorithms that maintain diverse, high-performing solutions — MAP-Elites, QD optimization.
| Prio | Paper | Authors | Venue | Year | ID |
|---|---|---|---|---|---|
| !!! | Evolvability ES: Scalable and Direct Optimization of Evolvability | Gajewski, Clune, Stanley, Lehman | GECCO | 2019 | 1907.06077 |
| !! | Illuminating search spaces by mapping elites | Mouret, Clune | arXiv | 2015 | 1504.04909 |
| !! | Quality Diversity: A New Frontier for Evolutionary Computation | Pugh, Soros, Stanley | Front Robot AI | 2016 | — |
| !! | Evolving Populations of Diverse RL Agents with MAP-Elites | Flageat et al. | arXiv | 2023 | 2303.12803 |
| ! | Scaling MAP-Elites to Deep Neuroevolution | Colas et al. | GECCO | 2020 | — |
| !! | EvoPrompt: Connecting LLMs with Evolutionary Algorithms | — | ICLR | 2024 | — |
| !! | Promptbreeder: Self-Referential Self-Improvement via Prompt Evolution | — | ICML | 2024 | — |
| !! | GEPA: Reflective Prompt Evolution Can Outperform RL | — | arXiv | 2025 | — |
Kategorie 2: Quality-Diversity / Evolvierbarkeit
Algorithmen, die diverse, hochperformante Lösungen pflegen — MAP-Elites, QD-Optimierung.
| Prio | Paper | Autoren | Venue | Jahr | ID |
|---|---|---|---|---|---|
| !!! | Evolvability ES: Scalable and Direct Optimization of Evolvability | Gajewski, Clune, Stanley, Lehman | GECCO | 2019 | 1907.06077 |
| !! | Illuminating search spaces by mapping elites | Mouret, Clune | arXiv | 2015 | 1504.04909 |
| !! | Quality Diversity: A New Frontier for Evolutionary Computation | Pugh, Soros, Stanley | Front Robot AI | 2016 | — |
| !! | Evolving Populations of Diverse RL Agents with MAP-Elites | Flageat et al. | arXiv | 2023 | 2303.12803 |
| ! | Scaling MAP-Elites to Deep Neuroevolution | Colas et al. | GECCO | 2020 | — |
| !! | EvoPrompt: Connecting LLMs with Evolutionary Algorithms | — | ICLR | 2024 | — |
| !! | Promptbreeder: Self-Referential Self-Improvement via Prompt Evolution | — | ICML | 2024 | — |
| !! | GEPA: Reflective Prompt Evolution Can Outperform RL | — | arXiv | 2025 | — |
Category 3: Self-Evolving Agents (Surveys)
Comprehensive surveys covering what, when, and how agents can self-improve.
| Prio | Paper | Authors | Venue | Year | ID |
|---|---|---|---|---|---|
| !!! | A Survey of Self-Evolving Agents | Gao et al. | arXiv | 2025 | 2507.21046 |
| !!! | A Comprehensive Survey of Self-Evolving AI Agents | Fang et al. (CN) | arXiv | 2025 | 2508.07407 |
| !!! | Foundation Agents: Brain-Inspired to Evolutionary, Collaborative, Safe | MetaGPT + Mila + 20 inst. (CN) | arXiv | 2025 | — |
| !! | 智能体技术和应用研究报告 (Agent Tech Report) | CAICT + Huawei (CN) | Report | 2025 | — |
Kategorie 3: Self-Evolving Agents (Surveys)
Umfassende Surveys: Was, wann und wie Agents sich selbst verbessern können.
| Prio | Paper | Autoren | Venue | Jahr | ID |
|---|---|---|---|---|---|
| !!! | A Survey of Self-Evolving Agents | Gao et al. | arXiv | 2025 | 2507.21046 |
| !!! | A Comprehensive Survey of Self-Evolving AI Agents | Fang et al. (CN) | arXiv | 2025 | 2508.07407 |
| !!! | Foundation Agents: Brain-Inspired to Evolutionary, Collaborative, Safe | MetaGPT + Mila + 20 inst. (CN) | arXiv | 2025 | — |
| !! | 智能体技术和应用研究报告 (Agent Tech Report) | CAICT + Huawei (CN) | Report | 2025 | — |
Category 4: Evolving Agent Workflows (Core Topic)
The core of our topic — papers that evolve agent workflows automatically.
| Prio | Paper | Authors | Venue | Year | ID |
|---|---|---|---|---|---|
| !!! | EvoFlow: Evolving Diverse Agentic Workflows On The Fly | Zhang, Chen, Wan et al. (CN) | arXiv | 2025 | 2502.07373 |
| !!! | EvoAgentX: Automated Framework for Evolving Agentic Workflows | Wang, Liu, Fang, Meng | EMNLP Demo | 2025 | 2507.03616 |
| !!! | Meta Context Engineering via Agentic Skill Evolution | Ye, He, Arak, Dong, Song | arXiv | 2026 | 2601.21557 |
| !!! | AgentFactory: Self-Evolving via Executable Subagent Accumulation | Zhang, Lu, Qian, He, Liu (CN) | arXiv | 2026 | 2603.18000 |
| !! | SEW: Self-Evolving Agentic Workflows for Code Generation | Liu, Fang et al. | arXiv | 2025 | 2505.18646 |
| !! | MermaidFlow: Safety-constrained Evolutionary Workflow Programming | — | arXiv | 2025 | — |
| !! | EvolveR: Self-Evolving LLM Agents through Experience-Driven Lifecycle | — | arXiv | 2025 | 2510.16079 |
| !! | AgentEvolver: Efficient Self-Evolving Agent System | — | arXiv | 2025 | 2511.10395 |
| !!! | Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents | Zhang, Hu, Lu, Lange, Clune (Sakana AI) | arXiv | 2025 | 2505.22954 |
| !! | EvoRL: Evolution of Agentic Workflows Using Reinforcement Learning | Attota, X | BigData Conf | 2025 | — |
| !! | Polymath: Self-Optimizing Agent with Dynamic Hierarchical Workflow | — | arXiv | 2025 | 2508.02959 |
| ! | Self-Organizing Agent Network for LLM-based Workflow Automation | — | arXiv | 2025 | 2508.13732 |
Kategorie 4: Evolving Agent Workflows (Kernthema)
Der Kern unseres Themas — Papers, die Agent-Workflows automatisch evolvieren.
| Prio | Paper | Autoren | Venue | Jahr | ID |
|---|---|---|---|---|---|
| !!! | EvoFlow: Evolving Diverse Agentic Workflows On The Fly | Zhang, Chen, Wan et al. (CN) | arXiv | 2025 | 2502.07373 |
| !!! | EvoAgentX: Automated Framework for Evolving Agentic Workflows | Wang, Liu, Fang, Meng | EMNLP Demo | 2025 | 2507.03616 |
| !!! | Meta Context Engineering via Agentic Skill Evolution | Ye, He, Arak, Dong, Song | arXiv | 2026 | 2601.21557 |
| !!! | AgentFactory: Self-Evolving via Executable Subagent Accumulation | Zhang, Lu, Qian, He, Liu (CN) | arXiv | 2026 | 2603.18000 |
| !! | SEW: Self-Evolving Agentic Workflows for Code Generation | Liu, Fang et al. | arXiv | 2025 | 2505.18646 |
| !! | MermaidFlow: Safety-constrained Evolutionary Workflow Programming | — | arXiv | 2025 | — |
| !! | EvolveR: Self-Evolving LLM Agents through Experience-Driven Lifecycle | — | arXiv | 2025 | 2510.16079 |
| !! | AgentEvolver: Efficient Self-Evolving Agent System | — | arXiv | 2025 | 2511.10395 |
| !!! | Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents | Zhang, Hu, Lu, Lange, Clune (Sakana AI) | arXiv | 2025 | 2505.22954 |
| !! | EvoRL: Evolution of Agentic Workflows Using Reinforcement Learning | Attota, X | BigData Conf | 2025 | — |
| !! | Polymath: Self-Optimizing Agent with Dynamic Hierarchical Workflow | — | arXiv | 2025 | 2508.02959 |
| ! | Self-Organizing Agent Network for LLM-based Workflow Automation | — | arXiv | 2025 | 2508.13732 |
Category 5: Multi-Agent Self-Evolution
Systems where multiple agents co-evolve and improve each other.
| Prio | Paper | Authors | Venue | Year | ID |
|---|---|---|---|---|---|
| !!! | SEMAG: Self-Evolutionary Multi-Agent Code Generation | Peng, Hou, Zhu et al. (CN) | arXiv | 2026 | 2603.15707 |
| !!! | SAGE: Multi-Agent Self-Evolution for LLM Reasoning | Peng, Zhu et al. (CN) | arXiv | 2026 | 2603.15255 |
| !! | OpenHospital: Arena for Evolving Collective Intelligence | Liu et al. (CN) | arXiv | 2026 | 2603.14771 |
| !! | Loosely-Structured Software: Evolution Entropy in MAS | Zhang, Zhou et al. (CN) | arXiv | 2026 | 2603.15690 |
| !! | MetaAgent: Automatic MAS Construction via FSM | — | ICML | 2025 | — |
| !! | MAS-GPT: Training LLMs to Construct Multi-Agent Systems | — | ICML | 2025 | — |
| !! | MAS-ZERO: Designing MAS with Zero Supervision | — | arXiv | 2025 | — |
| !! | Evolving Interpretable Constitutions for Multi-Agent Coordination | Kumar et al. | arXiv | 2026 | 2602.00755 |
| !!! | CoMAS: Co-Evolving Multi-Agent Systems via Interaction Rewards | Xue et al. (CN) | ICLR | 2026 | 2510.08529 |
| !! | EvoScientist: Multi-Agent Evolving AI Scientists for Scientific Discovery | Lyu et al. (CN) | arXiv | 2026 | 2603.08127 |
| !! | Learning to Deliberate: Meta-policy Collaboration for Agentic LLMs | — | arXiv | 2025 | 2509.03817 |
Kategorie 5: Multi-Agent Self-Evolution
Systeme, in denen mehrere Agents ko-evolvieren und sich gegenseitig verbessern.
| Prio | Paper | Autoren | Venue | Jahr | ID |
|---|---|---|---|---|---|
| !!! | SEMAG: Self-Evolutionary Multi-Agent Code Generation | Peng, Hou, Zhu et al. (CN) | arXiv | 2026 | 2603.15707 |
| !!! | SAGE: Multi-Agent Self-Evolution for LLM Reasoning | Peng, Zhu et al. (CN) | arXiv | 2026 | 2603.15255 |
| !! | OpenHospital: Arena for Evolving Collective Intelligence | Liu et al. (CN) | arXiv | 2026 | 2603.14771 |
| !! | Loosely-Structured Software: Evolution Entropy in MAS | Zhang, Zhou et al. (CN) | arXiv | 2026 | 2603.15690 |
| !! | MetaAgent: Automatic MAS Construction via FSM | — | ICML | 2025 | — |
| !! | MAS-GPT: Training LLMs to Construct Multi-Agent Systems | — | ICML | 2025 | — |
| !! | MAS-ZERO: Designing MAS with Zero Supervision | — | arXiv | 2025 | — |
| !! | Evolving Interpretable Constitutions for Multi-Agent Coordination | Kumar et al. | arXiv | 2026 | 2602.00755 |
| !!! | CoMAS: Co-Evolving Multi-Agent Systems via Interaction Rewards | Xue et al. (CN) | ICLR | 2026 | 2510.08529 |
| !! | EvoScientist: Multi-Agent Evolving AI Scientists for Scientific Discovery | Lyu et al. (CN) | arXiv | 2026 | 2603.08127 |
| !! | Learning to Deliberate: Meta-policy Collaboration for Agentic LLMs | — | arXiv | 2025 | 2509.03817 |
Category 6: Multi-Agent Science (Attribution, Failure, Benchmarks)
Attribution, failure modes, benchmarks — making multi-agent research rigorous.
| Prio | Paper | Authors | Venue | Year | ID |
|---|---|---|---|---|---|
| !!! | Towards a Science of Collective AI (Collaboration Gain Γ) | — | arXiv | 2026 | 2602.05289 |
| !! | MultiAgentBench: Collaboration and Competition of LLM agents | Zhu et al. | ACL | 2025 | — |
| !! | Why Do Multi-Agent LLM Systems Fail? (MAST) | Cemri et al. | arXiv | 2025 | 2503.13657 |
| !! | Multi-Agent Collaboration Mechanisms: A Survey | Tran et al. | arXiv | 2025 | 2501.06322 |
| ! | Lessons from 2025 on Agents and Trust | Google Cloud | Report | 2025 | — |
| !! | AgenTracer: Who Is Inducing Failure in LLM Agentic Systems? | — | arXiv | 2025 | 2509.03312 |
| !! | Rethinking the Value of Multi-Agent Workflow: Strong Single Agent Baseline | — | arXiv | 2026 | 2601.12307 |
| ! | AgentDropoutV2: Optimizing Information Flow in MAS via Pruning | — | arXiv | 2026 | 2602.23258 |
| ! | SC-MAS: Cost-Efficient MAS with Edge-Level Heterogeneous Collaboration | — | arXiv | 2026 | 2601.09434 |
Kategorie 6: Multi-Agent Science (Attribution, Failure, Benchmarks)
Attribution, Fehlermodi, Benchmarks — Multi-Agent-Forschung rigoros machen.
| Prio | Paper | Autoren | Venue | Jahr | ID |
|---|---|---|---|---|---|
| !!! | Towards a Science of Collective AI (Collaboration Gain Γ) | — | arXiv | 2026 | 2602.05289 |
| !! | MultiAgentBench: Collaboration and Competition of LLM agents | Zhu et al. | ACL | 2025 | — |
| !! | Why Do Multi-Agent LLM Systems Fail? (MAST) | Cemri et al. | arXiv | 2025 | 2503.13657 |
| !! | Multi-Agent Collaboration Mechanisms: A Survey | Tran et al. | arXiv | 2025 | 2501.06322 |
| ! | Lessons from 2025 on Agents and Trust | Google Cloud | Report | 2025 | — |
| !! | AgenTracer: Who Is Inducing Failure in LLM Agentic Systems? | — | arXiv | 2025 | 2509.03312 |
| !! | Rethinking the Value of Multi-Agent Workflow: Strong Single Agent Baseline | — | arXiv | 2026 | 2601.12307 |
| ! | AgentDropoutV2: Optimizing Information Flow in MAS via Pruning | — | arXiv | 2026 | 2602.23258 |
| ! | SC-MAS: Cost-Efficient MAS with Edge-Level Heterogeneous Collaboration | — | arXiv | 2026 | 2601.09434 |
Category 7: Memory & Cognition Evolution
How agents build and curate long-term knowledge — the heredity mechanism.
| Prio | Paper | Authors | Venue | Year | ID |
|---|---|---|---|---|---|
| !! | Memory as Asset: Human-centric Memory Management | Pan, Huang, Yang | arXiv | 2026 | 2603.14212 |
| !! | AutoAgent: Evolving Cognition + Elastic Memory | Wang et al. (CN) | arXiv | 2026 | 2603.09716 |
| !! | TheraAgent: Multi-Agent with Self-Evolving Memory | Chen et al. (CN) | arXiv | 2026 | 2603.13676 |
| !! | RetroAgent: Retrospective Dual Intrinsic Feedback | Zhang, Liu et al. (CN) | arXiv | 2026 | 2603.08561 |
| ! | Steve-Evolving: Open-World Embodied Self-Evolution | Xie et al. (CN) | arXiv | 2026 | 2603.13131 |
| !!! | MemEvolve: Meta-Evolution of Agent Memory Systems | — | arXiv | 2025 | 2512.18746 |
| !! | Self-Evolving Distributed Memory Architecture for Scalable AI | — | arXiv | 2026 | 2601.05569 |
Kategorie 7: Memory & Cognition Evolution
Wie Agents Langzeit-Wissen aufbauen und kuratieren — der Vererbungsmechanismus.
| Prio | Paper | Autoren | Venue | Jahr | ID |
|---|---|---|---|---|---|
| !! | Memory as Asset: Human-centric Memory Management | Pan, Huang, Yang | arXiv | 2026 | 2603.14212 |
| !! | AutoAgent: Evolving Cognition + Elastic Memory | Wang et al. (CN) | arXiv | 2026 | 2603.09716 |
| !! | TheraAgent: Multi-Agent with Self-Evolving Memory | Chen et al. (CN) | arXiv | 2026 | 2603.13676 |
| !! | RetroAgent: Retrospective Dual Intrinsic Feedback | Zhang, Liu et al. (CN) | arXiv | 2026 | 2603.08561 |
| ! | Steve-Evolving: Open-World Embodied Self-Evolution | Xie et al. (CN) | arXiv | 2026 | 2603.13131 |
| !!! | MemEvolve: Meta-Evolution of Agent Memory Systems | — | arXiv | 2025 | 2512.18746 |
| !! | Self-Evolving Distributed Memory Architecture for Scalable AI | — | arXiv | 2026 | 2601.05569 |
Category 8: Self-Testing & Quality Gates
Evaluation-as-architecture — agents that test and verify their own outputs.
| Prio | Paper | Authors | Venue | Year | ID |
|---|---|---|---|---|---|
| !! | Automated Self-Testing as Quality Gate for LLM Apps | Maiorano | arXiv | 2026 | 2603.15676 |
| ! | Learning to Negotiate: Multi-Agent Collective Value Alignment | Anantaprayoon et al. | arXiv | 2026 | 2603.10476 |
Kategorie 8: Selbst-Test & Quality Gates
Evaluation-as-Architecture — Agents, die ihre eigenen Outputs testen und verifizieren.
| Prio | Paper | Autoren | Venue | Jahr | ID |
|---|---|---|---|---|---|
| !! | Automated Self-Testing as Quality Gate for LLM Apps | Maiorano | arXiv | 2026 | 2603.15676 |
| ! | Learning to Negotiate: Multi-Agent Collective Value Alignment | Anantaprayoon et al. | arXiv | 2026 | 2603.10476 |
Category 9: Chinese-Language Sources
Relevant work from Chinese institutions, often with distinct approaches.
Kategorie 9: Chinesisch-sprachige Quellen
Relevante Arbeiten aus chinesischen Institutionen, oft mit eigenen Ansätzen.
Statistics
- Total Papers: 60+
- !!! (Must-Read): 19
- Chinese involvement: ~25 papers (marked CN)
- 2026 Papers: 20+
- Open Repos: EvoAgentX, EvoFlow, CoMAS, Darwin Gödel Machine, pyribs, QDax, MultiAgentBench
- Last updated: 2026-03-22 (added 13 papers via Semantic Scholar citation graph + CJK scan)
Statistik
- Total Papers: 60+
- !!! (Must-Read): 19
- Chinesische Beteiligung: ~25 Papers (CN markiert)
- 2026 Papers: 20+
- Offene Repos: EvoAgentX, EvoFlow, CoMAS, Darwin Gödel Machine, pyribs, QDax, MultiAgentBench
- Zuletzt aktualisiert: 22.03.2026 (13 Papers via Semantic Scholar Citation Graph + CJK-Scan hinzugefügt)