Discovering Algorithms with Computational Language Processing

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Bourdais, Theo, Gnanasekaran, Abeynaya, Owhadi, Houman, Sahai, Tuhin
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909685038186496
author Bourdais, Theo
Gnanasekaran, Abeynaya
Owhadi, Houman
Sahai, Tuhin
author_facet Bourdais, Theo
Gnanasekaran, Abeynaya
Owhadi, Houman
Sahai, Tuhin
contents Algorithms are the engine for reproducible problem-solving. We present a framework automating algorithm discovery by conceptualizing them as sequences of operations, represented as tokens. These computational tokens are chained using a grammar, enabling the formation of increasingly sophisticated procedures. Our ensemble Monte Carlo tree search (MCTS) guided by reinforcement learning (RL) explores token chaining and drives the creation of new tokens. This methodology rediscovers, improves, and generates new algorithms that substantially outperform existing methods for strongly NP-hard combinatorial optimization problems and foundational quantum computing approaches such as Grover's and Quantum Approximate Optimization Algorithm. Operating at the computational rather than code-generation level, our framework produces algorithms that can be tailored specifically to problem instances, not merely classes.
format Preprint
id arxiv_https___arxiv_org_abs_2507_03190
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Discovering Algorithms with Computational Language Processing
Bourdais, Theo
Gnanasekaran, Abeynaya
Owhadi, Houman
Sahai, Tuhin
Artificial Intelligence
Data Structures and Algorithms
Machine Learning
es: 68T05, 68T20, 68Q12, 90C27
I.2.6; I.2.8; F.2.2; F.1.2; G.2.1
Algorithms are the engine for reproducible problem-solving. We present a framework automating algorithm discovery by conceptualizing them as sequences of operations, represented as tokens. These computational tokens are chained using a grammar, enabling the formation of increasingly sophisticated procedures. Our ensemble Monte Carlo tree search (MCTS) guided by reinforcement learning (RL) explores token chaining and drives the creation of new tokens. This methodology rediscovers, improves, and generates new algorithms that substantially outperform existing methods for strongly NP-hard combinatorial optimization problems and foundational quantum computing approaches such as Grover's and Quantum Approximate Optimization Algorithm. Operating at the computational rather than code-generation level, our framework produces algorithms that can be tailored specifically to problem instances, not merely classes.
title Discovering Algorithms with Computational Language Processing
topic Artificial Intelligence
Data Structures and Algorithms
Machine Learning
es: 68T05, 68T20, 68Q12, 90C27
I.2.6; I.2.8; F.2.2; F.1.2; G.2.1
url https://arxiv.org/abs/2507.03190