EMI-Group/evox

What it solves

EvoX addresses the performance bottlenecks and scalability issues associated with traditional evolutionary computation. It provides a high-performance, distributed framework that allows researchers and developers to run complex evolutionary algorithms (EAs) across CPUs and GPUs, achieving significant speedups (over 100x) compared to non-accelerated implementations.

How it works

Built to be compatible with PyTorch, EvoX uses a tailored programming model to accelerate the execution of evolutionary processes. It organizes its functionality into a hierarchical architecture that supports single-objective and multi-objective optimization, as well as advanced tasks like neuroevolution and hyperparameter optimization. The framework integrates with physics engines (such as Brax) and provides a specialized .exv format for efficient real-time data streaming and visualization.

Who it’s for

It is designed for researchers and engineers working in evolutionary computation, reinforcement learning, and neuroevolution who need to scale their experiments across multiple nodes or hardware accelerators.

Highlights

• Massive Algorithm Library: Includes over 50 evolutionary algorithms, covering Differential Evolution, Evolution Strategy, Particle Swarm Optimization, and various multi-objective approaches.
• Extensive Benchmarking: Provides 100+ benchmark problems spanning numerical challenges and RL environments.
• Hardware Acceleration: Supports heterogeneous computing on CPUs and GPUs with seamless distributed workflow scaling.
• PyTorch Integration: Fully compatible with the PyTorch ecosystem, making it easier to integrate with neural network models for neuroevolution.