PLAN: Parallel Liquid-Inspired Approximation Network for Efficient Representation Learning in Flexible Job Shop Scheduling

ReLA: Representation Learning and Aggregation for Scalable Job Scheduling with Reinforcement Learning

Large-scale job scheduling is a classic problem in computing systems and industrial operations, where complex workloads, workflows, or ordered job operations must be assigned to computing nodes or machines under resource, precedence, and availability constraints. Existing solvers can provide useful reference solutions, but their search cost is often too high for time-sensitive scheduling. Recent reinforcement-learning (RL) schedulers offer faster inference, yet many rely on limited state representations, which can weaken action scoring as scheduling instances scale. In this paper, we propose ReLA, an RL scheduler built on structured representation learning and aggregation. ReLA learns intra-entity representations using self-attention and convolution, captures inter-entity operation–machine interactions using cross-attention, and aggregates multi-scale representations for parallel actor-based scoring of feasible actions. Experiments on synthetic and public scheduling benchmarks show that ReLA achieves the best makespan in most tested settings. On small and medium instances, ReLA achieves a 7.3% average optimality gap and reduces the state-of-the-art (SOTA) baseline gap by 13.0%. On large instances with at least a hundred jobs, ReLA reduces the SOTA gap by 78.6%, with an average gap of 2.1%. These results demonstrate ReLA’s effectiveness for scalable and runtime-efficient scheduling over large action spaces.

Bidirectional job-shop scheduling with graph-theoretic features: a graph attention network and proximal policy optimization approach

A novel DRL-based approach that integrates bidirectional scheduling with graph-theoretic features to effectively solve JSSP is introduced, which achieves superior performance compared to traditional heuristics and classic DRL methods, while maintaining competitive results against recent state-of-the-art approaches.

Solving the hybrid action space problem in edge computing: a graph-aware reinforcement learning approach for container management

A Two-Layer Multi-Agent Deep Reinforcement Learning Framework for Flexible Job-Shop Scheduling with Multiple Batch-Processing Machines

Advancements in intelligent manufacturing require solutions to the conventional flexible job-shop scheduling problem (FJSP) to accommodate increasingly intricate constraints, particularly in the semiconductor and electronic component sectors, where batch-processing machines (BPMs) significantly intensify scheduling complexity. To address this challenge, this study formulates an extended FJSP with multiple BPMs and proposes an end-to-end two-layer multi-agent deep reinforcement learning framework. Job and machine agents perform decentralized action mapping, while workshop states are encoded using a heterogeneous disjunctive graph and a dual-graph attention network. Unlike standard FJSP learning methods that primarily address operation–machine decisions, the proposed framework jointly models machine assignment, operation sequencing, variable-length batch formation, and BPM allocation within a unified policy, with a pointer network-based batching agent and an equipment-selection agent that handle batch-processing decisions under feasibility masking. The framework was validated using plant-derived production data and multi-scale synthetic instances. Numerical results show that the proposed method achieves competitive performance across the tested batching and standard-FJSP settings. In standard-FJSP comparisons, relative performance was scenario-dependent: DANIEL performed better in S1, whereas both proposed variants outperformed all comparators in S2. These results support the framework as an effective scheduling approach for deterministic FJSP with BPMs and indicate cross-scale generalization across evaluated instances.

HARTS: Efficient Agentic Reinforcement Learning for Hybrid-Attention Models over Arbitrary Rollout Trees

HARTS is the first system to demonstrate arbitrary-rollout-tree prefix-sharing speedups on a real hybrid-attention model, and its numerical differences are comparable to baseline self-rerun variation, and its reward trend is similar to the baseline over the first 120 steps of SWE-bench training.

Preference-Guided Multi-Policy Optimization for Flexible Job Shop Scheduling

PGMPO is proposed, a novel learning framework consisting of a simple but effective multi-policy modeling approach that allows a single network to represent multiple decision-makers, and a preference-driven model optimization method that effectively guides policies to learn diverse and specialized problem-solving strategies without the need for explicit reward functions.