Benchmarking Classical and Quantum Reinforcement Learning Algorithms with JAX

hyperlax is a unified JAX-based framework for benchmarking and accelerating both classical and quantum reinforcement learning (RL). It provides a high-throughput environment for comparing machine-learning algorithms under identical computational settings, enabling reproducible and fair performance evaluations across model types such as multilayer perceptron (MLP), tensorized neural network and parametrized quantum circuit (PQC).

Key features

• Unified interface for classical, quantum, and tensor-network RL algorithms

• Batched hyperparameter exploration with vectorized execution

• Modular configuration system with strongly-typed dataclasses

• Compatible with HPC clusters through Singularity containers
  
  

Use cases

• Comparative benchmarking of classical vs. quantum RL agents

• Large-scale hyperparameter optimization using Optuna or random/QMC sampling