The Transformer Network for the Traveling Salesman Problem
Institute for Pure & Applied Mathematics (IPAM) via YouTube
Overview
Explore a 30-minute conference talk on applying the Transformer network to solve the Traveling Salesman Problem (TSP). Delve into Xavier Bresson's research from Nanyang Technological University, Singapore, presented at the Deep Learning and Combinatorial Optimization 2021 event. Learn about the adaptation of the Transformer architecture, originally developed for natural language processing, to tackle this classic combinatorial optimization challenge. Discover how reinforcement learning and beam search decoding are utilized to achieve improved performance over recent learned heuristics, with impressive optimal gap results for TSP50 and TSP100. Gain insights into the potential of deep learning to develop better heuristics for NP-hard problems, potentially revolutionizing approaches to combinatorial challenges in industry.
Syllabus
Introduction
Deep Learning
Architecture
Comparison
Coding
Discussion
Taught by
Institute for Pure & Applied Mathematics (IPAM)