Tuesday, August 28 2018
12:15pm - 1:15pm
Pettit Microelectronics Building, Room 102 A&B
For more information:

Allie McFadden

Communications Officer

allie.mcfadden@cc.gatech.edu

 

Add To My Calendar
ML@GT Talk — Bryan McCann, Salesforce

The Machine Learning Center at Georgia Tech (ML@GT) is excited to welcome Bryan McCann from Salesforce to campus for a ML@GT Talk.

For scheduling information, contact Mark Riedl at riedl@cc.gatech.edu

Please RSVP by Tuesday, August 27th.

Title
The Natural Language Decathlon: Multitask Learning as Question Answering

Abstract
Deep learning has improved performance on many natural language processing (NLP) tasks individually. However, general NLP models cannot emerge within a paradigm that focuses on the particularities of a single metric, dataset, and task. 

We introduce the Natural Language Decathlon (decaNLP), a challenge that spans ten tasks:
question answering, machine translation, summarization, natural language inference, sentiment analysis, semantic role labeling, zero-shot relation extraction, goal-oriented dialogue, semantic parsing, and commonsense pronoun resolution. We cast all tasks as question answering over a context. Furthermore, we present a new Multitask Question Answering Network (MQAN) jointly learns all tasks in decaNLP without any task-specific modules or parameters in the multitask setting. MQAN shows improvements in transfer learning for machine translation and named entity recognition, domain adaptation for sentiment analysis and natural language inference, and zero-shot capabilities for text classification. We demonstrate that the MQAN's multi-pointer-generator decoder is key to this success and performance further improves with an anti-curriculum training strategy.
Though designed for decaNLP, MQAN also achieves state of the art results on the WikiSQL semantic parsing task in the single-task setting.

We release code for procuring and processing data, training and evaluating models, and reproducing all experiments for decaNLP. 

Bio

Bryan McCann is a Senior Research Scientist at Salesforce. He focuses on transfer learning and multitask learning for natural language processing. Most recently, Bryan proposed the Natural Language Decathlon (decaNLP) and a Multitask Question Answering Network to tackle all ten tasks in decaNLP. Before decaNLP, he showed that the intermediate representations, or context vectors (CoVe), of machine translation systems carry information that aids learning in question answering and text classification systems.
Prior to working at Salesforce, Bryan studied at Stanford University, where he completed a B.S and M.S in Computer Science as well as a B.A in Philosophy. 

Click images in enlarge.