(本文轉(zhuǎn)自智能算法研究中心官方網(wǎng)站:https://www2.scut.edu.cn/huanghan/2024/1227/c9791a573754/page.htm)
學(xué)術(shù)報(bào)告一
報(bào)告時(shí)間:2024年12月30日(星期一)下午14:30
報(bào)告地點(diǎn):華南理工大學(xué)軟件學(xué)院B8報(bào)告廳
主持人:黃翰 教授
報(bào)告題目:Evolution of Heuristics: Towards Efficient Automated Algorithm Design Using Large Language Models
報(bào)告摘要:
Heuristics are widely used for dealing with complex search and optimization problems. However, manual design of heuristics can be often very labour extensive and requires rich working experience and knowledge. In this talk, I will introduce Evolution of Heuristic (EoH), an evolutionary paradigm that leverages both Large Language Models (LLMs) and evolutionary search for Automatic Heuristic Design (AHD). EoH represents the ideas of heuristics in natural language, termed thoughts. They are then translated into executable codes by LLMs. The evolution of both thoughts and codes in an evolutionary search framework makes it very effective and efficient for generating high-performance heuristics. Experiments on three widely studied combinatorial optimization benchmark problems demonstrate that EoH outperforms commonly used handcrafted heuristics and other recent AHD methods including FunSearch proposed by google deepmind.
報(bào)告人:Chair Professor Qingfu Zhang
個(gè)人簡(jiǎn)介:
Qingfu Zhang is a Chair Professor of Computational Intelligence with the Department of Computer Science, City University of Hong Kong. His is an IEEE fellow. His main research interests include evolutionary computation, optimization, metaheuristic, machine learning and their applications. He leads the Optimization and learning Group in CityU. His MOEA/D algorithms have been widely researched and used in industry. He has been listed as a highly cited researcher in computer science for 8 times.
學(xué)術(shù)報(bào)告二
報(bào)告時(shí)間:2024年12月30日(星期一)下午15:30
報(bào)告地點(diǎn):華南理工大學(xué)軟件學(xué)院B8報(bào)告廳
主持人:黃翰 教授
報(bào)告題目:面向大規(guī)模路徑問題的神經(jīng)組合優(yōu)化方法
報(bào)告摘要:
神經(jīng)組合優(yōu)化 (NCO) 旨在直接從數(shù)據(jù)中學(xué)習(xí)一個(gè)能夠直接求解復(fù)雜組合優(yōu)化問題,,如旅行商問題 (TSP) 和容量受限車輛路徑問題 (CVRP),,的神經(jīng)網(wǎng)絡(luò)?,F(xiàn)有的 NCO 模型在小規(guī)模問題實(shí)例上取得了良好的表現(xiàn),,但無法泛化求解大規(guī)模問題。本報(bào)告將系統(tǒng)地回顧這些現(xiàn)有的 NCO 方法并介紹它們的基本原理,。本報(bào)告還將總結(jié)現(xiàn)有模型大規(guī)模泛化能力差的一些可能原因,。此外,本報(bào)告將介紹一種新的NCO模型,,輕型編碼器和重型解碼器 (LEHD),。將詳細(xì)介紹該模型的結(jié)構(gòu)、訓(xùn)練方案和推理策略,。之后,,我將介紹一個(gè)更具一般性的分而治之 (UDC) 框架,可用于解決的范圍更廣的組合優(yōu)化問題,。
報(bào)告人: 王振坤 副研究員
個(gè)人簡(jiǎn)介:
王振坤,,IEEE高級(jí)會(huì)員,、廣東省全驅(qū)系統(tǒng)理論與技術(shù)重點(diǎn)實(shí)驗(yàn)室副主任、南方科技大學(xué)自動(dòng)化與智能制造學(xué)院助理教授,、博士生導(dǎo)師,研究方向?yàn)槿斯ぶ悄芘c調(diào)度優(yōu)化,,以第一/通訊作者身份在IEEE匯刊,、ICML、NeurIPS等國(guó)際高水平期刊和會(huì)議上發(fā)表論文40余篇(其中IEEE匯刊論文21篇,、CCF A類會(huì)議論文12篇),,獲2023 年度中國(guó)仿真學(xué)會(huì)自然科學(xué)二等獎(jiǎng)(國(guó)家一級(jí)學(xué)會(huì),1/7),、2023 年度廣東省自然科學(xué)二等獎(jiǎng)(2/5),、第八屆全國(guó)青年人工智能創(chuàng)新創(chuàng)業(yè)大會(huì)創(chuàng)新組一等獎(jiǎng)(1/3)、第六屆智能優(yōu)化與調(diào)度學(xué)術(shù)會(huì)議青年科學(xué)家獎(jiǎng),、2024年度SWEVO最佳副編輯獎(jiǎng),、華為公司火花獎(jiǎng)等榮譽(yù);主持國(guó)自然面上等科研項(xiàng)目8項(xiàng),,擔(dān)任三個(gè)國(guó)際期刊的副編輯,、IEEE計(jì)算智能學(xué)會(huì)深圳分會(huì)學(xué)生事務(wù)主席、中國(guó)人工智能學(xué)會(huì)青年工作委員會(huì)委員,、中國(guó)仿真學(xué)會(huì)智能仿真優(yōu)化與調(diào)度專委會(huì)委員以及多個(gè)國(guó)際會(huì)議的程序委員會(huì)委員,。
主辦單位:廣東省計(jì)算機(jī)學(xué)會(huì)軟件工程專業(yè)委員會(huì)
承辦單位:華南理工大學(xué)軟件學(xué)院
附:華工進(jìn)校報(bào)備流程
被訪人:凌蓮芬
手機(jī)號(hào)碼:13570468639
被訪人單位類型:學(xué)院
被訪人單位:軟件學(xué)院