Korali
High-performance framework for uncertainty quantification, optimization and reinforcement learning.
Korali is a high-performance framework for Bayesian UQ, optimization, and reinforcement learning. Korali’s multi-language interface allows the execution of any type of computational model, either sequential or distributed (MPI), C++ or Python, and even pre-compiled/legacy applications. Korali’s execution engine enables scalable sampling on large-scale HPC systems.
Korali provides a simple interface that allows users to easily describe statistical / deep learning problems and choose the algorithms to solve them. The framework can easily be extended to describe new problems or test new experimental algorithms on existing problems.
For more information, read: S. Martin, D. Waelchli, G. Arampatzis, A. E. Economides and P. Karnakov, P. Koumoutsakos, “Korali: Efficient and Scalable Software Framework for Bayesian Uncertainty Quantification and Stochastic Optimization”. arXiv 2005.13457. Zurich, Switzerland, March 2021. [PDF].
Usage
Run with Docker: docker run -it cselab/korali:latest
Documentation: https://korali.readthedocs.io/
Website: https://www.cse-lab.ethz.ch/korali/
Contact us
The Korali Project is developed and maintained by
Sergio Miguel Martin, martiser at ethz.ch
Daniel Waelchli, wadaniel at ethz.ch
Georgios Arampatzis, garampat at ethz.ch
Pascal Weber, webepasc at ethz.ch
Frequent contributors: Fabian Wermelinger, Lucas Amoudrouz, Ivica Kicic
User Manual
- Bayesian Inference
- Design
- Hierarchical Bayesian
- Integration
- Optimization
- Model Propagation
- Propagation of Uncertainty
- Reaction: Simulating the evolution of reactants
- Deep Reinforcement Learning
- Upswing (Python)
- Mountain Cart (Python)
- Cartpole (Python)
- Multi-Agent Reinforcement Learning for Flow Control
- Multi-Agent Reinforcement Learning with VMAS: A Vectorized Multi-Agent Simulator for Collective Robot Learning
- Study Case: Reinforcement Learning on Artificial Bacterial Flagella (ABF) in 2D
- Multi-Agent Reinforcement Learning on Active Particles
- Sampling
- Supervised Learning