Prerequisites: to follow the hands-on session on HPC please have your ZIH login at hand.
A: Guided hands-on
- In the session, you'll receive ready-to-use notebooks to acquaint yourself with fundamental HPC features. Following this, you'll be equipped to start with your own research question using HPC. No prerequisites are required.
B: Consultancy with your own code/data
- Please bring your own pipeline/problem to work on, with the assistance of our HPC experts. We aim to help you with setting up your software stack, starting with distributed training, setting up an ML Ops workflow, investigating performance, and many more.
C: OmniOpt - Efficient Parallel Hyperparameter Optimization on HPC
-In this session, learn to optimize hyper-parameters (HPs) crucial for neural networks and other machine-learning methods efficiently. Explore OmniOpt, a specialized tool for TU Dresden's HPC cluster, designed to streamline HP optimization for various problems. Leveraging HPCs powerful GPUs, CPUs, and storage ensures swift handling of even large problems. We will cover (i) a brief HP optimization overview, (ii) hands-on OmniOpt usage with a practical example, and (iii) result evaluation using OmniOpt's toolbox. You are encouraged to bring your own code for optimization.
Link: https://scads.ai/transfer-2/verfuegbare-software-dienste-en/omniopt/