This hands-on course will take you from 0 to 100 in Deep Learning with Keras. Our aim is to teach the fundamentals of deep learning with Convolutional Neural Networks (CNN) based on modern techniques using the Keras API and the Tensorflow backend. By the end participants will know how to build deep learning models, how to train them, what to avoid during training, what to check during training and how to perform model inference, especially for image based problems. We hope participants will then go out and apply these methods to their own problems and use cases.
The core curriculum is planned from Tuesday (August 22) to Thursday night (August 24) to take place at the MPI CBG campus, Dresden, Germany. As the agenda is currently being prepared please check-in from time to time. There is the latent plan to have voluntary session before (introduction to python) and after the workshop (hackday to work on the challenges you have). This is why the workshop currently blocks the entire week.
All participants are expected to bring their laptop. During the workshop, a uniform access to GPU-enabled workstations or servers will be provided that hold the software stack used. Thus, your laptop is not required to hold a mobile GPU or alike. All participants are expected to have a solid understanding of fundamentals of linear algebra as well as programming.
The workshop admission fee amounts to € 100 per participant. Every successfull applicant is required to bring a poster to the workshop that describes their current scientific challenge that they would like to solve with Deep Learning. Posters have to be sent in 1 week prior to the workshop.
Here are our confirmed instructors (in no order of preference):
Nico Hofmann earned his PhD (Dr. rer. nat.) in 2016 from the Technische Universität Dresden (Dresden, Germany) with his work on intraoperative medical image analysis. The thesis was supervised by Sen.-Prof. Dr.-Ing. Uwe Petersohn. While pursuing his PhD, he worked as research assistant for the Clinical Sensoring and Monitoring lab of the TU Dresden in close corporation with the University Hospital Dresden. During this time, he developed statistical machine learning methods to compensate motion artefacts and advanced unsupervised semiparametric regression models to recognize weak patterns originating from evoked neuronal activity of the human brain. The latter is used as novel tool for intraoperative visualization of neural activity in order to discriminate healthy from tumor tissue. Additionally, he and his students developed novel approaches for multimodal 2D-3D image registration and fusion in Amira. Recently, Nico designed deep neural networks that efficiently solve inverse problems of non-linear and partial differential equations as well as for pattern recognition in multivariate time-series data by learning low-dimensional representations of the data.
Kashif Rasul earned his B.Sc. with Honours from Monash University in Melbourne, Australia, followed by his Ph.D. in 2010 from the Free University of Berlin in the area of Differential Geometry and PDEs. He completed his thesis under the supervision of Prof. Klaus Ecker. Prior to finishing his Ph.D. he worked at the Max Planck Institute for Gravitational Physics (Albert Einstein Institute) on the Cactus Framework, an open-source, problem-solving environment designed for scientists and engineers. Kashif also worked as a software developer on Amira, a 3D scientific visualisation framework. Since this time, he has been the cofounder of two startups in the area of Geospatial Databases and Crowdsourced Logistics. Currently he is working as a Research Scientist at Zalando Research in Berlin. Kashif recently began his postdoctoral research at the Free University of Berlin in the Databases group, lead by Prof. Agnès Voisard. He is also employed as a teacher there. Kashif is passionate about high performance computing and has presented at various conferences, including Nvidia’s GTC and Strange Loop. He is an avid contributor to open source projects via github.com/kashif.
Matias Valdenegro is currently a PhD Student at Heriot-Watt University (Edinburgh, Scotland, UK) in his final year, funded by the Marie Curie ITN "Robocademy" and advised by Prof David Lane. He has previously received a Master's Degree in Autonomous Systems from Hochschule Bonn-Rhein-Sieg (Sankt Augustin, Germany), and an Bachelor of Science and Computer Engineering Degree from Universidad Tecnologica Metropolitana (Santiago, Chile). His Master Thesis is based on his work at the Fraunhofer Institute for Intelligent Analysis and Information Systems. His thesis research received the AFCEA Studienpreis (1st place) in 2015. Previous to his research career in Europe he worked for four years in the Chilean software development industry. Matias' doctoral research is about object detection and recognition in sonar images, specifically perception for marine debris (garbage) recovery, using deep neural networks. His broad interests are in Robot Vision, Underwater Robotics and Machine Learning.
And here are our sponsors: