Semantic Representation Learning

This event has ended.

The provisional timetable for the workshop is as follows:

11:00-11:10 Welcome

11:10-12:20 Shay Cohen (University of Edinburgh)

12:20-13:20 Lunch

13:20-14:30 Sebastian Riedel (UCL)

14:30-15:40 Thomas Lukasiewicz (University of Oxford)

15:40-16:00 Coffee break

16:00-17:00 Angelika Kimmig (Cardiff University)

Representation learning is the task of converting data from the raw form in which it is given into a form which is more suitable for machine learning models. Given the popularity of deep learning, the most common approach is to learn a mapping from data items onto a fixed-dimensional vector. Such vector representations are commonly used in image processing and natural language processing in particular. A particular challenge, however, is to learn vector representations which are, in some sense, semantically meaningful. On the one hand, this is critical for enabling machine learning models which are interpretable, a property that is quickly becoming critical in many applications. On the other hand, semantically meaningful representations are needed to incorporate existing domain knowledge (e.g. provided by a domain expert or available from some knowledge base), and thus to inject knowledge into machine learning models.