On September 22, Oliver Dürr, Beate Sick and Elvis Murina hosted the 2nd Deep Learning Day at Zurich University of applied science (ZHAW) in Winterthur. As a start, the organizers gave a review of the deep learning approach and convolutional neural networks in general. The presentation illustrated the impact on many artificial intelligence tasks such as semantic segmentation, natural language processing, and image caption generation. The subsequent presentation revealed the challenging aspects of deep learning in medical imaging and showed that there is considerable room for improvement in medical applications of deep neural networks.
Two in-depth tutorials from NVidia boosted the technical knowledge of participants. First, MXNet, a flexible and efficient library for deep learning was introduced. The framework can be used for deep learning together with Amazon Web Services (AWS). Second, the effect of choosing a suitable loss function for human heart volume estimation from a time-series of MRI data was presented.
The afternoon saw a presentation on Long Short-Term Memory (LSTM) Recurrent Neural Networks (RNNs) for sequence-to-sequence learning. Conversational agents can be trained end-to-end to generate genuine replies in a dialogue with interaction with humans. Then, the attendees learned how to forecast time-series using LSTMs in Keras during a second tutorial by NVidia.
The workshop concluded with a very interesting poster session where participants presented many practical applications of deep learning: For example, medical applications included stroke detection, intracranial aneurysms localization, medical audio processing, and lung cancer classification. Another group of posters presented newspaper articles segmentation, text mining, and text comprehension. The interactive format of poster presentations led to a vibrant exchange of ideas.
In summary, this 2nd Deep Learning Day was a great event. The participants not only gained a lot of knowledge, but also established new connections. Attendees with research and industrial interests with a variety of scientific expertise provided a unique opportunity for networking and knowledge exchange. The detailed content of the workshop is available via this link: