Skip to main content

PRACE Deep Learning Workshop at PDC

Thor Wikfeldt, KTH

On the 20th and 21st of March, PDC organized a workshop on deep learning sponsored by PRACE. As could be expected, based on the current surge of enthusiasm surrounding deep learning methods, the workshop quickly filled up after registration was opened. The workshop was taught by two instructors from Finland, Markus Koskela and Mats Sjöberg, who both work at the CSC - IT Center for Science. Their extensive experience in developing, applying and teaching deep learning methods has enabled them to develop a comprehensive yet practical two-day introduction to deep learning - not an easy task considering the complexity of the subject! On the first day of the workshop, participants logged in to a CSC cloud service which provides access to Jupyter Notebooks with all the deep learning packages pre-installed, but no GPUs. On the second day, when more computationally demanding methods were explored, participants logged in to Tegner and ran their experiments on the K80 and K420 GPUs available on Tegner compute nodes. The lesson material mainly focused on convolutional neural networks (CNNs), recurrent neural networks (RNNs) and multi-layer perceptron (MLP) networks, along with important concepts such as supervised versus unsupervised learning, activation functions, backpropagation, automatic differentiation, pooling and dropout. Most of the exercises, analysing both image and text data, were based on using the Keras package with TensorFlow in the back end, but participants had the option to also try out the PyTorch package.

The course feedback we received showed that the participants were very happy with the workshop. PDC aims to deliver further workshops on machine learning and deep learning methods in the future, so keep an eye on the PDC website or subscribe to the pdc-announce mailing list if you are interested in attending!