Training course "Introduction to Deep Learning Models"
(Course no. 1142019 in the training programme 2019 of Forschungszentrum Jülich)
This course is fully booked.
Target audience: | Researchers who want to use deep-learning methods |
Contents: | |
Prerequisites: | Participants should be able to work on the Unix/Linux command line, have a basic level of understanding of batch scripts required for HPC application submissions, and have a minimal knowledge of probability, statistics, and linear algebra. |
Language: | This course is taught in English. |
Duration: | 3 days |
Date: | 21-23 May 2019, on 21 May 13:00-16:30 on 22 and 23 May 09:00-16:30 |
Venue: | Jülich Supercomputing Centre, Rotunda, building 16.4, room 301 |
Number of participants: | maximum 40, working in groups of 2 |
Instructors: | Prof. Dr. Morris Riedel, Dr. Gabriele Cavallaro, Dr. Jenia Jitsev, JSC |
Contact: | Dr. Gabriele Cavallaro Phone: +49 2461 61-3858 E-mail: g.cavallaro@fz-juelich.de |
Registration: | This course is fully booked. |
This course focuses on a recent machine learning method known as deep learning that emerged as a promising disruptive approach, allowing knowledge discovery from large datasets in an unprecedented effectiveness and efficiency. It is particularly relevant in research areas, which are not accessible through modelling and simulation often performed in HPC. Traditional learning, which was introduced in the 1950s and became a data-driven paradigm in the 90s, is usually based on an iterative process of feature engineering, learning, and modelling. Although successful on many tasks, the resulting models are often hard to transfer to other datasets and research areas.
This course provides an introduction into deep learning and its inherent ability to derive optimal and often quite generic problem representations from the data (aka ‘feature learning’). Concrete architectures such as Convolutional Neural Networks (CNNs) will be applied to real datasets of applications using known deep learning frameworks such as Tensorflow, Keras, or Torch. As the learning process with CNNs is extremely computational-intensive the course will cover aspects of how parallel computing can be leveraged in order to speed-up the learning process using general purpose computing on graphics processing units (GPGPUs). Hands-on exercises allow the participants to immediately turn the newly acquired skills into practice. After this course participants will have a general understanding for which problems CNN learning architectures are useful and how parallel and scalable computing is facilitating the learning process when facing big datasets.
This course is a training course of the DEEP-EST project.