Transformer-based Framework for Multi-objective Reinforcement Learning using Hierarchical Policies

Research Project

Project/Area Number	24K20843
Research Category	Grant-in-Aid for Early-Career Scientists
Allocation Type	Multi-year Fund
Review Section	Basic Section 61030:Intelligent informatics-related
Research Institution	Kyushu University
Principal Investigator	シュレスタマリサソット九州大学, システム情報科学研究院, 学術研究員 (80970607)
Project Period (FY)	2024-04-01 – 2027-03-31
Project Status	Granted (Fiscal Year 2024)
Budget Amount *help	¥4,680,000 (Direct Cost: ¥3,600,000、Indirect Cost: ¥1,080,000) Fiscal Year 2026: ¥780,000 (Direct Cost: ¥600,000、Indirect Cost: ¥180,000) Fiscal Year 2025: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000) Fiscal Year 2024: ¥2,990,000 (Direct Cost: ¥2,300,000、Indirect Cost: ¥690,000)
Keywords	Reinforcement Learning / Artificial Intelligence / Transformers / Deep Learning
Outline of Research at the Start	Reinforcement Learning (RL) has demonstrated remarkable proficiency in various domains. However, there is still much to be done to enable RL agents to strategize long-term policies and navigate through environments with multiple reward signals. This study aims to forge a comprehensive framework wherein RL agents can efficiently acquire modular hierarchical policies for multi-objective optimization challenges by leveraging the attention mechanism in Transformer networks. This will also enable the distillation of specialized solutions from overarching policies for specific application scenarios.