REVIEW 2 cited by
Ranked Reward: Enabling Self-Play Reinforcement Learning for Combinatorial Optimization
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Adversarial self-play in two-player games has delivered impressive results when used with reinforcement learning algorithms that combine deep neural networks and tree search. Algorithms like AlphaZero and Expert Iteration learn tabula-rasa, producing highly informative training data on the fly. However, the self-play training strategy is not directly applicable to single-player games. Recently, several practically important combinatorial optimisation problems, such as the travelling salesman problem and the bin packing problem, have been reformulated as reinforcement learning problems, increasing the importance of enabling the benefits of self-play beyond two-player games. We present the Ranked Reward (R2) algorithm which accomplishes this by ranking the rewards obtained by a single agent over multiple games to create a relative performance metric. Results from applying the R2 algorithm to instances of a two-dimensional and three-dimensional bin packing problems show that it outperforms generic Monte Carlo tree search, heuristic algorithms and integer programming solvers. We also present an analysis of the ranked reward mechanism, in particular, the effects of problem instances with varying difficulty and different ranking thresholds.
Forward citations
Cited by 2 Pith papers
-
One4Many-StablePacker: An Efficient Deep Reinforcement Learning Framework for the 3D Bin Packing Problem
One deep RL model for 3D bin packing generalizes to unseen bin dimensions and enforces stability constraints, via a weighted loading-rate/height-difference reward and entropy-controlled PPO.
-
DOPPLER: Dual-Policy Learning for Device Assignment in Asynchronous Dataflow Graphs
DOPPLER trains two cooperating neural policies, one that orders graph operations and one that maps them to GPUs, to reduce execution time in asynchronous work-conserving multi-GPU systems.
Discussion (0). Continue with ORCID to comment.