Speaker: Riccardo Bonetto, Date: 11th of March 2016, Time: 14:30, Room: 326
Abstract. Recurrent neural networks are an extension of the more utilized feedforward neural networks. While feedforward neural networks have exhibited very good performance in tasks like handwritten characters classification and pattern recognition, their performance is not as good when dealing with timeseries forecasting, speech recognition, and, in general, with tasks utilizing datasets with variable size and exhibiting dynamic temporal behavior. Recurrent neural networks allow to address this issue by introducing the concept of memory. In this talk recurrent neural networks will be introduced and the main challenges that must be faced when implementing these networks will be presented.