Home // eKNOW 2025, The Seventeenth International Conference on Information, Process, and Knowledge Management // View article

Heuristic Search Using Language Models and Reinforcement Learning

Authors:
Carolina Carvalho
Paulo Quaresma

Keywords: Heuristic Optimization; Reinforcement Leaning; Language Model; Task Semantic Segmentation; Artificial Neural Network.

Abstract:
This article extends the applicability domain of language models to problems in which candidate solutions can be expressed by an encoded integer sequence. Considering this sequence, language models can work in the neural machine translation setting and bring their optimization power to the heuristic search technique. Reinforcement Learning (RL) is applied to Language Models (LM), whether char-level or word-level is used as a basic framing. In order to stabilize the learning, several approaches are explored such as functional and architecture decoupling. The framework is then applied to two combinatorial problems, namely the Traveling Salesman Problem benchmark and Neural Architecture Search, used to generate an hierarchical (tree-based) text classifier where the blocks are inspired by the InceptionV1 architecture. The decoupling results are this paper’s main contribution, easing the RL plus LM stabilization requirements and opening the resolution domain beyond Markov-Decision-Processes, to non-causal normative heuristic problems such as Neural Architecture Search (NAS).

Pages: 1 to 12

Copyright: Copyright (c) IARIA, 2025

Publication date: May 18, 2025

Published in: conference

ISSN: 2308-4375

ISBN: 978-1-68558-272-2

Location: Nice, France

Dates: from May 18, 2025 to May 22, 2025