← 全部范式 / 10

搜索、仿真与科学计算

搜索、仿真、自对弈和科学结构预测结合了神经网络与结构化推理。

12 篇论文 第 10 个，共 10 个计算范式

Search, simulation, and science compute

英文原文文件：README.md

设备/设置

CPU/GPU/TPU actor-learner 系统、搜索服务器、模拟器和科学管线，将学习模型与显式探索或优化相结合。

瓶颈

环境模拟、树搜索、自博弈数据生成、长程信用分配和科学结构搜索，比单次前向计算更显著地决定了整体成本。

适配的方法

DQN replay、AlphaGo/AlphaZero/MuZero 搜索、AlphaStar/OpenAI Five 联赛自博弈，以及 AlphaFold 式学习势能或结构预测，让学习适配模拟和科学优化。

变得过时或不再中心的方法

纯监督模仿、手工游戏启发式和经典采样管线，在学习模型可大规模引导搜索或模拟后退居次要地位。

代表性论文

排名 年份 论文 优先级 状态
111 2015 Human-level control through deep reinforcement learning 5 downloaded / read_complete
112 2016 Mastering the game of Go with deep neural networks and tree search 5 downloaded / read_complete
113 2017 Mastering the game of Go without human knowledge 5 downloaded / read_complete
114 2021 Highly accurate protein structure prediction with AlphaFold 5 downloaded / read_complete
115 2018 A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play 4 downloaded / read_complete
116 2019 Mastering Atari, Go, chess and shogi by planning with a learned model 4 downloaded / read_complete
117 2024 Accurate structure prediction of biomolecular interactions with AlphaFold 3 4 downloaded / read_complete
118 2019 Grandmaster level in StarCraft II using multi-agent reinforcement learning 3 html_saved_no_open_pdf / read_complete
119 2019 Dota 2 with Large Scale Deep Reinforcement Learning 3 downloaded / read_complete
120 2020 Improved protein structure prediction using potentials from deep learning 3 downloaded / read_complete
127 2025 Gold-medalist Performance in Solving Olympiad Geometry with AlphaGeometry2 4 downloaded / read_complete
128 2025 AlphaEvolve: A coding agent for scientific and algorithmic discovery 5 downloaded / read_complete

开放问题

  • 比较游戏/模拟计算与科学推理管线，而非将两者都简化为泛化深度学习。

相关论文 12

2015 Human-level control through deep reinforcement learning
2016 Mastering the game of Go with deep neural networks and tree search
2017 Mastering the game of Go without human knowledge
2021 Highly accurate protein structure prediction with AlphaFold
2018 A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play
2019 Mastering Atari, Go, chess and shogi by planning with a learned model
2024 Accurate structure prediction of biomolecular interactions with AlphaFold 3
2019 Grandmaster level in StarCraft II using multi-agent reinforcement learning
2019 Dota 2 with Large Scale Deep Reinforcement Learning
2020 Improved protein structure prediction using potentials from deep learning
2025 Gold-medalist Performance in Solving Olympiad Geometry with AlphaGeometry2
2025 AlphaEvolve: A coding agent for scientific and algorithmic discovery