-
Notifications
You must be signed in to change notification settings - Fork 26
Home
LLM for Algorithm Design
基于大语言模型与进化计算的自动算法设计平台
LLM4AD_Next 是一个开源、模块化的自动算法设计平台,把大语言模型(LLM)与进化计算结合起来,自动完成算法的设计与优化。过去设计一个算法往往需要人工建模、反复编码与调参;而在 LLM4AD_Next 中,你只需用一句自然语言描述需求,平台便能自动完成算法设计、代码生成、评估与进化优化的完整闭环,并支持网页版全程可视化、内置 VSCode 随手改随手跑。
- 🌟 GitHub: Optima-CityU/LLM4AD_Next
我们会持续发布介绍平台能力与实战案例的文章,欢迎按语言查看:
- 🇨🇳 中文 → 推文合集(中文)
- 🇬🇧 English → Article Collection (EN)
两个合集涵盖同一系列,包括:
- 新功能发布 —— 集成 FunSearch / ReEvo / EoH 等主流方法,覆盖在线装箱、TSP、流水车间调度等案例
- 记忆功能上新 —— 接入华为诺亚方舟自研的 MindMemOS,让算法设计从"单次搜索"走向"持续积累",提供任务/项目/全局三层记忆与自动召回/手动指定两种调用方式
- No.1 · 符号回归 —— 从科学数据中发现数学公式
- No.2 · 组合优化 —— 用大模型自动设计启发式算法(TSP / CVRP),同源的「大模型 + 进化计算」范式曾夺得 2026 CVRPLib BKS 全球挑战赛冠军
- 平台发布 —— 「设计算法?我们只需要一句话」
LLM4AD_Next is an open-source, modular platform that combines Large Language Models (LLMs) with evolutionary computation to automatically design and optimize algorithms. Instead of manual modeling, endless coding, and repeated tuning, you describe what you need in one sentence of natural language, and the platform runs the full loop of design, code generation, evaluation, and evolutionary optimization — with a fully visualized web interface and a built-in VSCode for editing and running on the fly.
The two collections linked above cover the same article series:
- New Feature: Method Integration — FunSearch / ReEvo / EoH and other mainstream methods, covering online bin packing, TSP, flow-shop scheduling, and more
- New Feature: Memory Function — Powered by MindMemOS from Huawei Noah's Ark Lab, turning algorithm design from "one-shot search" into "continuous accumulation" with three-layer memory and two invocation modes
- No.1 · Symbolic Regression — Discovering mathematical formulas from scientific data
- No.2 · Combinatorial Optimization — Automatically designing heuristic algorithms with LLMs (TSP / CVRP); the same "LLM + evolutionary computation" paradigm won the 2026 CVRPLib BKS Global Challenge
- Platform Release — "Designing algorithms? All it takes is one sentence"