Tag
GAE introduces a framework combining graph neural networks, reinforcement learning, and LLM fine-tuning to overcome bottlenecks in evolutionary program search, achieving state-of-the-art performance on symbolic regression for complex nonlinear oscillator systems.
Introduces an evolutionary neural architecture search framework (EvoTS) for discovering task-adaptive Transformer-like models for multivariate time-series forecasting. The approach uses a modular genome representation and achieves competitive performance on ETT benchmark datasets.
This paper introduces Evolution Fine-Tuning (EFT), a mid-training paradigm that teaches LLMs to evolve solutions across optimization tasks by converting evolutionary search trajectories into supervision. The authors construct the Finch Collection dataset of 156K trajectories across 371 tasks and show that fine-tuned models generalize to held-out tasks and match state-of-the-art performance on several benchmarks.
MiniMax open-sourced MaxProof, a test-time scaling framework for LLM mathematical proofs, and released a companion paper. The framework uses an evolutionary search mechanism to enable the M3 model to achieve gold-medal scores on both the IMO 2025 and USAMO 2026 test sets.
Evoflux uses evolutionary search at inference time to repair failed tool workflows for compact language models, boosting execution feasibility significantly over fine-tuning methods.
This paper proposes an evolutionary framework inspired by parallel tempering that uses multi-temperature sampling and information exchange to improve the diversity and quality of scientific hypotheses generated by large language models, demonstrated across molecular, equation, and algorithm discovery.
SePO (Self-Evolving Prompt Optimization) proposes a self-referential prompt agent that optimizes both task agents' system prompts and its own system prompt through an evolutionary search, outperforming Manual-CoT, TextGrad, and MetaSPO across five benchmarks including AIME'25, ARC-AGI-1, and GPQA.
Introduces BES (Bidirectional Evolutionary Search), a search framework for LLMs that combines forward candidate evolution with backward goal decomposition to improve sampling on hard reasoning problems during post-training and inference.
This paper presents optimize_anything, a universal LLM-based optimization system for text artifacts that achieves state-of-the-art results across diverse tasks including agent architecture discovery, scheduling, CUDA kernel generation, and packing, demonstrating general-purpose text optimization.
This paper introduces DIO-Agent, a discovery agent that synthesizes programs from input-output behavior using LLM-guided evolutionary search with a transformation priority premise to avoid dead ends. Experiments show it outperforms traditional methods and baselines on a new IO2CodeBench benchmark.
This paper investigates execution-grounded automated AI research by building an automated executor that implements LLM-generated ideas and runs experiments. It shows that execution-guided evolutionary search can find methods that significantly outperform baselines in both pre-training and post-training tasks.
Introduces Persona Policies (PPol), a plug-and-play control layer that uses LLM-driven evolutionary program search to generate diverse, human-like user personas for evaluating LLM agents. Achieves 33–62% fitness gains over baseline, with human-likeness rated at 80.4%, and improves agent robustness with +17% task success.
The paper introduces PACEvolve++, a reinforcement learning framework that improves test-time policy adaptation for evolutionary search agents by decoupling hypothesis generation from execution.
Large-scale study of 15 LLMs across 8 tasks reveals that optimization success hinges on maintaining localized search trajectories rather than initial problem-solving ability or solution novelty.