Main lecture: Bálint Gyires-Tóth (BME-TMIT):  Deep Learning Based Time Series Modeling on Steroids 30’

Deep learning based sequence modeling has shown superior results to previous analytical and machine learning methods. Originally recurrent neural networks, including Long Short-Term Memory (LSTM) architecture, were designed for such purposes. However, convolutional neural networks (CNNs) have achieved outstanding results in many application scenarios lately (eg. speech synthesis and recognition, natural language processing). The structure of CNN allows us to be able to effectively train very deep networks (with more than 20-40, or even 100 layers), which is beneficial as deep architectures usually result in better performance than shallow models. The presentation will introduce deep learning based sequence modeling and discuss some of the state of the art methods as well.

Complementary lecture: Csongor Pilinszki-Nagy (BME-TMIT): Hierarchical Temporal Memory Models Alternative to Time Series Models? 15’

Complementary opinion: Garrett Mindt (CEU): Very Deep Neural Networks - Epistemic Implications 15’

Q&A 20’