Tid: Torsdag 8 december 2016, kl. 15:00-17:00
Plats: C307, Södra huset, Frescati

Abstract

A big part of computational linguistics deals with the automatic analysis of natural language using machine learning techniques. Current approaches rely heavily on a set of such techniques commonly referred to as ‘deep learning’.  When posing a natural language problem to a computer, researchers previously had to manually define all potentially relevant features. Deep learning allows for the automatic learning of these features, greatly simplifying the process from problem to fully fledged machine learning system. The majority of papers at conferences in our field currently include the use of deep learning, whereas just three years ago barely anyone had heard of it. It is therefore no surprise that this sudden transition of the field has been characterised as a tsunami, appearing (seemingly) out of nowhere.

In this seminar, I’ll give a linguist-friendly introduction to deep learning and how I apply it in my research. I’ll attempt to cover a range of questions, going from ‘What is machine learning?’ to ‘Why would you use a bidirectional LSTM when a simple perceptron should suffice?’. In doing so, I hope to give everyone a good idea what deep learning is and what it can do, where the field of computational linguistics is currently headed, as well as what I aim to cover in my PhD thesis.

Mer information om Johannes Bjerva

Hjärtligt välkomna!

Ljuba Veselinova