Scaling up Stochastic Algorithms through non-reversibility, parallelization and adaptness – SuSa
Scaling up Stochastic algorithms
through non-reversibility, parallelization and adaptness
Objectives
Both statistical physics and Bayesian statistical inference revolve around probabilistic modeling and relies on description based on intractable integrals, which are sampled through Monte Carlo algorithms. As size and complexity of problems increase, these high-dimensional problems require more efficient stochastic methods, yet still simple and robust. They offer an opportunity for rich and multidisciplinary collaborations, from computational physics to probability theory and Bayesian statistics. The SuSa project propose to bring together physicists, probabilists and computer scientists, so as to develop innovative algorithms, based on the most recent advances in Monte Carlo and learning methods. Both theoretical and practical issues will be addressed, from the production of precise analysis of the underlying stochastic processes to the development of numerical solutions for parallelization and computational complexity reduction and applications to large-scale dataset in physics.
Development and generalisation of sampling methods by non-reversivle stochastic processes.
Automation and optimization by adaptive and learning methods.
Applications in different fields of physics.
--
--
--
Both statistical physics and Bayesian statistical inference revolve around probabilistic modeling and relies on description based on intractable integrals, which are sampled through Monte Carlo algorithms. As size and complexity of problems increase, these high-dimensional problems require more efficient stochastic methods, yet still simple and robust. They offer an opportunity for rich and multidisciplinary collaborations, from computational physics to probability theory and Bayesian statistics. The SuSa project propose to bring together physicists, probabilists and computer scientists, so as to develop innovative algorithms, based on the most recent advances in Monte Carlo and learning methods. Both theoretical and practical issues will be addressed, from the production of precise analysis of the underlying stochastic processes to the development of numerical solutions for parallelization and computational complexity reduction and applications to large-scale dataset in physics.
Project coordination
Manon Michel (LABORATOIRE DE MATHEMATIQUES BLAISE PASCAL)
The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.
Partner
LMBP LABORATOIRE DE MATHEMATIQUES BLAISE PASCAL
Help of the ANR 193,147 euros
Beginning and duration of the scientific project:
September 2020
- 48 Months