AI Summary: Finding Increasingly Large Extremal Graphs with AlphaZero and Tabu Search

This paper studies the problem of finding graphs that maximize the number of edges, while avoiding short cycles. It formulates graph generation as a reinforcement learning task, and compares methods like AlphaZero and tabu search. A key finding is that using a curriculum - building larger graphs from good smaller graphs - significantly improves performance. The work makes progress on an open problem in extremal graph theory.

Anian Ruoss (@anianruoss) / X

Petar Veličković on LinkedIn: #icml2023

Petar Veličković - CatalyzeX

Satheesh Kumar V na LinkedIn: Synechron CEO's Winning Formula: Agile, ethical, and culturally strong

Nearly 100 Mila-affiliated scientific papers accepted at NeurIPS 2023 - Mila

Petar Veličković posted on LinkedIn

Anian Ruoss (@anianruoss) / X

LinkedInのPetar Veličković: Our Learning on Graphs Conference tutorial on Neural Algorithmic Reasoning…

Explainable artificial intelligence: an analytical review - Angelov - 2021 - WIREs Data Mining and Knowledge Discovery - Wiley Online Library

Adam Zsolt Wagner

State of AI Report: Compute Index V2 update!

Finding Increasingly Large Extremal Graphs with AlphaZero and Tabu Search

PDF] Proving Theorems using Incremental Learning and Hindsight Experience Replay