Understanding and Improving Deep Learning (FCS Research Seminar)

FCS RESEARCH SEMINAR SERIES | Monday, May 10 | 12:30 p.m. AST | See MS Teams link below

CSCI 6999 Eligible

Understanding and Improving Deep Learning 
Dr. Cecilia Summers
PhD student, the University of Auckland, New Zealand

Deep learning has been very successful over the past few years, having been successfully applied to tasks in computer vision, natural language processing, and biomedicine. However, most of these successes require a large amount of data, which is one of the biggest problems in deep learning today. The first part of this talk will be about a nonlinear type of mixed-example data augmentation, which is an approach that generates synthetic training data to reduce this requirement. The second part of the talk examines the neural network optimization process itself, studying the effects of nondeterminism and its surprising relation to the instability of optimization, including two approaches for reducing the effects of instability.

BIO
Cecilia Summers is a graduating PhD student from the University of Auckland, New Zealand. Her research focuses on understanding and improving core elements of deep learning, ranging from augmentation applied to data during training, normalization layers in neural networks, and nondeterminism and instability in optimization. During her PhD she has done a research internship at Apple, and before her PhD she did a MS in Applied Mathematics.

Time

Starts:
Ends:

Location

Microsoft Teams meeting

Join on your computer or mobile app

Click here to join the meeting

Or call in (audio only)

+1 437-703-4646,,987956972#   Canada, Toronto

Phone Conference ID: 987 956 972#

Find a local number | Reset PIN

Dalhousie University sits on the ancestral and unceded territory of the Mi'kmaq. We are all Treaty people.

Learn More | Help | Meeting options