Séminaire Samovar présenté par Jesse Read, le 24 mai 2018 à 14h00 en G10

Le séminaire est annulé pour des raisons personnelles.

Quand: le jeudi 24 mai 2018, à 14h00
Où: en G10, à Télécom SudParis (Evry)

Title: Models and Applications for Multi-Output and Structured-Output Prediction

Abstract: There is an increasingly prevalent need for machine learning models to be able to make predictions for multiple output variables simultaneously. For example, multiple labels may be applied to an image, a text document can be assigned to multiple categories, multiple diagnoses associated with a patient, or an anomaly may be detected across several dimensions through time and feature space. There are also connections to areas as diverse as recommender systems and reinforcement learning. This scenario of multiple outputs is challenging with regard to the extra complexity involved – both algorithmic complexity in modelling dependence, and increased memory and running time demands. But it can also be an advantage; learning to predict one output can help in predicting another. This talk will outline the main challenges, and some important developments to tackle these challenges.

Bio: Jesse Read obtained his PhD in Computer Science from the University of Waikato, New Zealand, in 2010. Following this, he has carried out postdoctoral research in the Universidad Carlos III of Madrid, Aalto University in Finland, and Télécom ParisTech. Since 2016 he is an assistant professor at École Polytechnique.