Tehran Institute for Advanced Studies (TEIAS)

/ Active and Multitask Learning Approaches to Low-Resource Neural Machine Translation __ Gholam Reza Haffari

Talk

Active and Multitask Learning Approaches
to Low-Resource Neural Machine Translation

September 2, 2020
(12 Shahrivar, 1399)

Venue

This Talk is online

+982189174612

Dr. Gholam Reza Haffari

Associate Professor in Department of Data Science & AI of Monash University

Overview

Machine translation is being revolutionized by the introduction of neural models. However, it is challenging to train high-quality neural translation models in small training data conditions, as usually, these models have lots of parameters.

In this talk, I give an overview of our research to tackle this problem from different angles using active learning (AL) and multitask learning (MTL) approaches. Firstly, our AL approach automatically learns an acquisition (policy) function in a Markov decision process formulation of the AL problem. Secondly, our MTL approach automatically learns a training schedule using a Markov decision process formulation of the problem.

 

Biography

Reza’s research interests lie at the intersection of natural language processing and machine learning, including topics such as machine translation, parsing, structured prediction, deep learning, reinforcement learning, probabilistic graphical models, etc. He is an Associate Professor and the Director of the Vision and Language Group in the Department of Data Science and AI, Monash University, Australia. He associates himself with venues such as ACL, EMNLP, NAACL, and TACL.