Overcoming limited labeled data with semi-supervised learning and data augmentation: Applications in Echocardiography and Brain-Computer Interfaces
Abstract
Quals talk:
Machine learning has achieved stellar results in recent years, across many areas such as image and speech recognition. In many areas, machine learning, especially deep learning, can even achieve human- or super-human-level performance using large collections of labeled data.
Despite the tremendous success, there are many areas where high-quality labeled data is scarce and is prohibitively costly to obtain. The lack of labeled data has become a major barrier to the wide adoption of ML systems in these areas.
In the talk, I'm going to talk about two of my projects that faced this problem and how we attempt to address them. One is using ultrasound heart images to diagnose heart disease Aortic Stenosis (AS), another is using fNIRS data to classify subject mental workload.
Please join the meeting in Sococo VH 475, or Zoom.
Join Zoom Meeting: https://tufts.zoom.us/my/hzhzhz
PASSWORD: See colloquia email
Dial-in option not available.