High-Performance Uncertainty Quantification, Optimization, and Reinforcement Learning
Korali is a high-performance framework for uncertainty quantification, optimization, and deep reinforcement learning. Its engine provides support for large-scale HPC systems and a multi-language interface compatible with distributed computational models.
Gradient-free optimization algorithms that enable parallel sampling in HPC systems. Additionally, Korali implements many well-known gradient based optimization algorithms.
Bayesian Uncertainty Quantification
Bayesian inference methods for the inference computational models parameters and their associated uncertainty, based on experimental data. Additionally, it provides multiple likelihood models and prior distributions.
Deep Reinforcement Learning
Deep reinforcement learning algorithms for both discrete and continuous actions spaces. Multiple agents can run in parallel, distributed across different nodes.
The work distribution engine is optimized to fully harness computational resources of large-scale supercomputers, maximizing throughput and minimizing workload imbalance. Furthermore, Korali supports the execution of parallel (OpenMP, Pthreads), distributed (MPI, UPC++), and GPU-based (CUDA) models.
Experiments store their internal state regularly, allowing them to be resumed later in case of errors. The result of a resumed experiment is guaranteed to be exactly the same as one in a single run.
The framework is open-source, modular and extensible. Developers can integrate and test new modules which will automatically benefit from its parallel engine and fault-tolerance without additional effort.
The API is compatible with C/C++/Fortran and Python models. Additionally, the engine can sample from pre-compiled computational models.
Journal / Conference / arXiv Papers
Please use the following article to cite Korali:
S. Martin, D. Wälchli, 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. Zürich, Switzerland, March 2021.
D. Wälchli, S. Martin, A. Economides, L. Amoudruz, G. Arampatzis, X. Bian, P. Koumoutsakos, "Load Balancing in Large Scale Bayesian Inference". Proceedings of the Platform for Advanced Scientific Computing Conference (PASC2020). Geneva, Switzerland, June 2020.
G. Arampatzis, D. Wälchli, P. Weber, H. Rästas, and P. Koumoutsakos, "(μ, λ)-CCMA-ES for Constrained Optimization with an Application in Pharmacodynamics", Proceedings of the Platform for Advanced Scientific Computing Conference (PASC2019). Zürich, Switzerland, June 2019.[PDF]
Publications that used Korali
P. Karnakov, G. Arampatzis, I. Kiči ć, F. Wermelinger, D. Wälchli, C. Papadimitriou, P. Koumoutsakos "Data-driven inference of the reproduction number for COVID-19 before and after interventions for 51 European countries". Swiss Medical Weekly. Zürich, Switzerland, July 2020.
M. Chatzimanolakis, P. Weber, G. Arampatzis, D. Wälchli, P. Karnakov, I. Kiči ć, C. Papadimitriou, P. Koumoutsakos "Optimal Testing Strategy for the Identification of COVID-19 Infections". medRxiv 2020.07.20.20157818v2. Zürich, Switzerland, July 2020.