Researchers have introduced GraphAllocBench, a new benchmark designed to evaluate preference-conditioned policy learning (PCPL) in multi-objective reinforcement learning (MORL). This benchmark is built on a novel graph-based resource allocation environment called CityPlannerEnv, offering greater flexibility and scalability than previous PCPL benchmarks. GraphAllocBench includes customizable objectives, varied preference conditions, and complex Pareto fronts, along with new metrics like Proportion of Non-Dominated Solutions (PNDS) and Ordering Score (OS) to better assess PCPL algorithm performance. Experiments with existing PCPL algorithms and a new graph-aware PCPL-PPO baseline demonstrate the benchmark's utility in revealing failure modes and highlighting the potential of graph-based approaches, such as Graph Neural Networks (GNNs), for complex allocation tasks. AI
IMPACT Introduces a more scalable and flexible benchmark for multi-objective reinforcement learning, potentially accelerating research in preference-conditioned policy learning.
RANK_REASON The cluster contains a research paper introducing a new benchmark and methodology. [lever_c_demoted from research: ic=1 ai=1.0]
- CityPlannerEnv
- GraphAllocBench
- Graph Neural Networks
- Multi-Objective Reinforcement Learning
- Ordering Score
- Preference-Conditioned Policy Learning
- Proportion of Non-Dominated Solutions
- Zhiheng Jiang
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
