Topics

• Unsupervised machine learning in engineering and neuroscience: applications of ICA
  By: Mark V. Albert , Pavan Ramkumar , Anne Zhao , Jorge Yanar
  Experience Level: Intermediate
  Length: 90 Minutes
  Description: This talk with be a set of four short presentations guiding everyone through three applications of unsupervised machine learning. We begin with the classic cocktail party problem - how to automatically separate mixed voices recorded by microphones - presented by Jorge Yanar. This will be followed by a brief, intuitive explanation of the algorithm used to perform the task - Independent Components Analysis (ICA) described by Professor Mark Albert. Dr. Pavan Ramkumar will demonstrate how the same technique is applied to filter unwanted noise during neural recordings using EEG, and Anne Zhao will end with a demonstration of how the same coding strategy has led to insights in how the brain encodes sensory information in the early auditory and visual systems. Her demo will allow participants to develop their own simulated neural codes for processing visual images. The brief talks will consist of a Jupyter notebook for running code and displaying results. For those who wish to run the examples during the talk, it will be necessary to install Jupyter running Python version 3 (the Anaconda Python distribution is recommended to set this up). Links and setup instructions will be given prior to the talks for people to follow along on their laptops and try the examples on their own if desired.