Abstract
This seminar presents applications of deep neural networks to natural language processing. These deep architectures have enabled spectacular advances for many tasks, and a complete paradigm shift in the field. The seminar begins with a discussion of language models and suitable network architectures (recurrent neural networks, LSTM, GRU and transform). Piotr Bojanowski presents the optimization of the parameters of these models, as well as the approximations needed to apply these methods on a large scale. In a second part, he introduces machine translation systems and algorithmic improvements that have enabled us to build today's high-performance models.
