Mahjax: A GPU-Accelerated Mahjong Simulator for Reinforcement Learning in JAX

2026-05-20 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Expert, medium

Summary

Mahjax is a new GPU-accelerated Riichi Mahjong simulator developed in JAX, designed to facilitate reinforcement learning research, particularly for *tabula rasa* learning approaches. Riichi Mahjong is a multi-player, imperfect-information game with high stochasticity and complex state spaces, making it a challenging environment for AI. Unlike prior methods relying on supervised learning from human play logs, Mahjax supports learning from scratch by enabling large-scale rollout parallelization on GPUs. The simulator demonstrates impressive performance, achieving throughputs of up to 2 million steps per second under no-red rules and 1 million steps per second under red rules, utilizing eight NVIDIA A100 GPUs. It also includes a high-quality visualization tool for debugging. Experimental results confirm its utility, showing that agents can be effectively trained to improve their rank against baseline policies.

Key takeaway

For Machine Learning Engineers developing reinforcement learning agents for complex, imperfect-information games, Mahjax offers a critical tool. You can now accelerate *tabula rasa* learning research using its GPU-accelerated JAX environment, achieving millions of simulation steps per second. This enables faster iteration and more robust agent training, potentially leading to breakthroughs in domains mirroring real-world decision-making challenges. Consider integrating Mahjax for large-scale experimentation and benchmarking.

Key insights

Mahjax provides a high-throughput, GPU-accelerated JAX environment for *tabula rasa* reinforcement learning in complex imperfect-information games.

Principles

• Vectorized JAX environments enable massive RL parallelization.
• Imperfect-information games challenge *tabula rasa* learning.
• High-throughput simulation accelerates algorithm development.

Method

The article describes the implementation of Mahjax in JAX for vectorized Riichi Mahjong simulation, enabling GPU-accelerated parallel rollouts and agent training against baselines.

In practice

• Train RL agents for complex imperfect-information games.
• Debug agent behavior using the visualization tool.
• Benchmark *tabula rasa* learning algorithms at scale.

Topics

• Reinforcement Learning
• JAX
• GPU Acceleration
• Riichi Mahjong
• Imperfect Information Games
• Simulation Environments
• Tabula Rasa Learning

Code references

• aadimator/JaxARC
• Emerge-Lab/gpudrive

Best for: Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• Mahjax: A GPU-Accelerated Mahjong Simulator for Reinforcement Learning in JAX — Mahjax provides a GPU-accelerated JAX environment for tabula rasa reinforcement learning in complex imperfect-information games.
• daVinci-kernel: Co-Evolving Skill Selection, Summarization, and Utilization via RL for GPU Kernel Optimization — A reinforcement learning framework co-evolves skill selection, summarization, and utilization for GPU kernel optimization.
• LogNEO: A GPT-Neo Reinforcement Learning Framework for Accurate Real-Time Log Anomaly Detection — LogNEO leverages GPT-Neo and a novel position-aware RL reward scheme to achieve high-accuracy, real-time log anomaly detection.
• Value Bonuses using Ensemble Errors for Exploration in Reinforcement Learning — VBE uses ensemble errors to create value bonuses for first-visit optimism and deep exploration in RL.
• Building AI That Plays Atari Games Using Deep Reinforcement Learning (Python) — Deep Reinforcement Learning enables AI to learn complex game strategies through trial and error.
• Convex Markov Games and Beyond: New Proof of Existence, Characterization and Learning Algorithms for Nash Equilibria — Nash equilibria in General Utility Markov Games coincide with fixed points of projected pseudo-gradient dynamics.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.