This title appears in the Scientific Report :
2023
Vast parameters space exploration using L2L on ERAINS
Vast parameters space exploration using L2L on ERAINS
This tutorial features a session on a hyper-parameter optimization framework, called L2L, implementing the concept of learning to learn. The framework provides a selection of different optimization algorithms and makes use of multiple high performance computing back-ends (e.g., multi nodes, GPUs) to...
Saved in:
Personal Name(s): | Yegenoglu, Alper |
---|---|
van der Vlag, Michiel (Corresponding author) / Diaz, Sandra | |
Contributing Institute: |
Jülich Supercomputing Center; JSC |
Imprint: |
2023
|
Conference: | 32nd Annual Computational Neuroscience Meeting, Leipzig (Germany), 2023-07-15 - 2023-07-15 |
Document Type: |
Conference Presentation |
Research Program: |
SimLab Neuroscience Human Brain Project Specific Grant Agreement 3 Domain-Specific Simulation & Data Life Cycle Labs (SDLs) and Research Groups |
Publikationsportal JuSER |
This tutorial features a session on a hyper-parameter optimization framework, called L2L, implementing the concept of learning to learn. The framework provides a selection of different optimization algorithms and makes use of multiple high performance computing back-ends (e.g., multi nodes, GPUs) to do vast parameter space explorations in an automated and parallel fashion (Yegenoglu et al. 2022). During this session you will learn about the installation and use of the framework within the EBRAINS’ Collaboratory. Two use-cases will be explored during the tutorial: 1) A task learning example, in which spiking neural networks implemented in the neural spiking simulator NEST are optimized to solve the mountain car problem from theOpenAI Gym environment. The task encompasses a car which is placed at the bottom of a valley. The networks have to find a configuration to accelerate the car to reach the goal on top of the right hill. 2) The optimization of a TVB (Sanz Leon et al. 2013) simulation which has been used in a study for scale-integrated understanding of conscious and unconscious brain states and their mechanisms (Goldman et al. 2021). In this use-case, a set of 5 model variables will be automatically explored to find an optimal parametrization for synchronous and a-synchronous brain states. Participants will learn how to launch the simulations on FENIX’s High Performing Computing infrastructure. Additionally, participants will be able to deploy jobs on CPUs and GPUs making use of UNICORE from the EBRAINS’ Collaboratory. |