UNSUPERVISED CONCEPT MINING - Towards Self-learning Machines
|
Speaker (s):
LIN Wen-Yan Daniel
Research Associate Professor,
Osaka University
|
|
Date:
Time:
Venue:
|
|
January 15, 2019, Tuesday
10:00am - 11:00am
Meeting Room 5.1, Level 5
School of Information Systems
Singapore Management University
80 Stamford Road
Singapore 178902
|
|
ABSTRACT
Machine Learning is an increasing part of our daily lives. Today, it is used in a wide variety of roles ranging from vacuum cleaners, cars, bio-metrics and factory production. Occasionally, there is even talk about sentient robots taking over the world. Despite this optimism, current machine learning techniques have very limited capacity for self-learning.
We believe these limitations fundamentally stem from our limited understanding of properties of high dimensional space. Most interesting concepts can only be meaningfully described in high dimensions. However, high dimensions space creates significant computational and theoretical challenges for existing unsupervised learning frameworks. Our preliminary results indicate these challenges can be overcome. This creates the potential for end-to-end learning systems which parse raw data to discover information from raw data. My goal is to lay the mathematical and engineering foundations to make this a reality.
About the Speaker
Dr Lin Wen-Yan, Daniel graduated with a PhD from National University of Singapore in 2012. He is currently a Research Associate Professor at Osaka University. He has published many papers on unsupervised computer vision with topics ranging from structure-from-motion, feature correspondence and clustering. These are core problems which need to be solved if machines are to be able to think for themselves. However, to date, little progress has been made. Daniel’s talk today will focus on the reasons for the lack of progress and suggest possible research directions.
