Mathias Berglund

Mathias Berglund · Wed, 16.12.2015


I participated in two conferences: The conference on Neural Information Processing Systems, where I presented a poster on “Bidirectional Recurrent Neural Networks as Generative Models” and on “Semi-Supervised Learning with Ladder Networks”, and in the International Conference on Learning Representations, with a poster on “Techniques for Learning Binary Stochastic Feedforward Neural Networks”. The presentations went well, and especially the paper on ladder networks received very much attention from the participants. 

The topics covered in the conference on Neural Information Processing Systems were broad and insightful. In addition, with almost 4000 participants, I had the chance to discuss with a wide variety of practitioners in the field. Content-wise, the deep learning symposium and several of the workshops were the most insightful, as they presented the most novel ideas. From the conference papers, there were also many highly relevant papers for my research.

It was also useful to discuss with other participants, which in the short term was useful particularly with respect to changes in frameworks that are generally used in deep learning today. I learned a lot about the current trends and directions which has a direct impact on my day-to-day research work.

The International Conference on Learning Representations in general was very insightful, as it is basically only focused on deep learning, which is my research topic. In addition to several interesting techniques and new research ideas (for instance an optimizer I have used since the conference), there was a lot of networking in the conference.