PACEvolve: Enabling Progress-Aware Consistent Evolution
cs.NE, cs.LG
Submitted: 2026-01-15
Updated: 2026-09-10
Code: https://github.com/KellerJordan/modded-nanogpt
License: http://creativecommons.org/licenses/by/4.0/
The gist: Self-evolving agents powered by Large Language Models (LLMs) have emerged as a promising direction across diverse domains, including code optimization and scientific discovery, yet their core failure
Terminology
Abstract
Self-evolving agents powered by Large Language Models (LLMs) have emerged as a promising direction across diverse domains, including code optimization and scientific discovery, yet their core failure modes remain underexplored. Through a comprehensive empirical study, we identify that the model's reasoning becomes anchored to the local context of current hypotheses, overemphasizing low-level details while neglecting the broader search landscape. As a result, such agents become prone to context pollution and mode collapse, repeatedly revisiting flawed hypotheses and converging on suboptimal solutions. To address this challenge, we propose Progress-Aware Consistent Evolution (PACEvolve), a systematic framework for governing agent memory and search dynamics. PACEvolve overcomes these limitations through three key techniques: (1) Hierarchical Context Management (HCM), which structures historical trajectories while dynamically pruning branches to preserve a high-signal memory state; (2) Momentum-Based Backtracking (MBB), which monitors optimization progress to escape local minima; and (3) a self-adaptive Collaborative Evolution policy (CE) that balances intra-trajectory refinement with inter-trajectory knowledge transfer. By decoupling high-level idea generation from low-level code evaluation, PACEvolve maintains a global view of search momentum and achieves state-of-the-art results across complex evolutionary benchmarks.
Sources
- FLEX: Continuous Agent Evolution via Forward Learning from Experience
- IterResearch: Rethinking Long-Horizon Agents with Interaction Scaling
- Multi-Agent Evolve: LLM Self-Improve through Co-evolution
- GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning
- Pass@k Training for Adaptively Balancing Exploration and Exploitation of Large Reasoning Models
- Barbarians at the Gate: How AI is Upending Systems Research
- CodeEvolve: an open source evolutionary coding agent for algorithmic discovery and optimization
- SELA: Tree-Search Enhanced LLM Agents for Automated Machine Learning
- Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities
- Inference-Time Scaling for Complex Tasks: Where We Stand and What Lies Ahead
- Supervised Reinforcement Learning: From Expert Trajectories to Step-wise Reasoning
- How Much Backtracking is Enough? Exploring the Interplay of SFT and RL in Enhancing LLM Reasoning
- Training AI Co-Scientists Using Rubric Rewards
- ShinkaEvolve: Towards Open-Ended And Sample-Efficient Program Evolution
- Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains
- Long-context LLMs Struggle with Long In-context Learning
- Reinforcement Learning with Rubric Anchors
- CUDA-L1: Improving CUDA Optimization via Contrastive Reinforcement Learning
- Bootstrapping Task Spaces for Self-Improvement
- Policy Guided Tree Search for Enhanced LLM Reasoning
Related papers
- Evolutionary Ensemble of Agents
- Encoding and Decoding Temporal Signals with Spiking Bandpass Wavelets
- Large Language Models and Evolutionary Computation: A Critical Review of Bidirectional Interaction, Automated Algorithm Design, and Co-Adaptive Systems
- Learning Alzheimer's Disease Signatures by bridging EEG with Spiking Neural Networks and Biophysical Simulations
- Investigating Hyperparameter Optimization and Transferability for ES-HyperNEAT: A TPE Approach
- S-AI-Recursive: A Bio-Inspired and Temporal Sparse AI Architecture for Iterative, Introspective, and Energy-Frugal Reasoning