Title: 

Induction of Crosslingual Distributed Representations of Words

Abstract:

One appeal of unsupervised distributed representations of words is that
they can help alleviate data sparsity problems common in the supervised
learning set-up. In the first part of the talk, I will present our work on
jointly inducing and aligning word representations across a pair of
languages relying primarily on unannotated monolingual data. These
representations can be used for a number of crosslingual learning tasks,
where a learner can be trained on annotations present in one language and
directly applied to test data in another.

In the second part of the talk, I will give an overview of some of the the
problems we are tackling with Machine Learning at Amazon. The huge amount
of data that is generated and collected throughout the company is creating
a tremendous amount of opportunities for Machine Learning and Data
Science. Applications traditionally included recommender systems and
search. Now they also range from problems in computer vision and natural
language processing to problems in forecasting and robotics.

The first part of the talk is joint work with Ivan Titov and Binod
Bhattarai.