神经与进化计算：2025–2026 年度综述

An annual survey of the cs.NE arXiv category, covering research published from September 2025 – August 2026. Built on a close-reading set of 100 papers and organised into 32 chapters, it maps the year by the field's research sub-directions.

Book framework (share of the year's output):

• Evolutionary Algorithms for Numerical Optimization (83.6%) — This subfield focuses on enhancing evolutionary computation techniques, such as genetic algorithms and swarm intelligence, to solve complex, multi-objective, and continuous optimization problems with improved robustness and efficiency.
• Automated Heuristic Design via LLMs (7.8%) — This area explores the integration of Large Language Models with evolutionary strategies to automate the design and optimization of heuristics, prompts, and algorithms for solving diverse computational problems.
• Spiking Neural Networks and Neuromorphic Computing (5.4%) — This subfield addresses the development and hardware deployment of energy-efficient spiking neural networks, focusing on sparse computation, neuromorphic hardware compatibility, and efficient inference paradigms.
• Neural Architecture Search and Deep Learning (2.0%) — This direction investigates advanced neural architecture search methods, deep learning model training, and adaptive architectures to improve performance on complex machine learning tasks.
• Temporal Dynamics in Spiking Neuron Models (0.6%) — This niche area studies the temporal processing capabilities, biological plausibility, and dynamic behavior of spiking neurons and oscillatory networks for time-sensitive learning tasks.
• Evolution Strategies and Adaptive Control (0.4%) — This subfield focuses on robust evolution strategies, covariance matrix adaptation, and automated synthesis of control policies for adaptive systems and human-AI co-evolution scenarios.
• Theoretical Approximation in Neural Networks (0.1%) — This theoretical branch examines the mathematical properties of neural network functions, including approximation capabilities of ReLU networks and the design of forward-forward goodness functions.

Each sub-direction is covered across three chapters — how it framed its questions, the strength of its key findings, and its open debates and frontiers — so the reader gets a traceable picture of what this part of the field actually produced over the year.