Towards effective statistical-neural hybrid machine learning

PhD Thesis Proposal Defence


Title: "Towards effective statistical-neural hybrid machine learning"

by

Mr. Yuchen YAN


Abstract:

We will propose two paradigms for closely integrating statistical and neural 
machine learning that make machine learning more efficient: (1) A statistical 
model and a neural network model should be able to co-train such that 
improvements to the statistical model can improve the neural network model and 
improvements to neural network model can improve the statistical model. In this 
way, the two models can form a feedback loop. (2) A neural network model should 
not just take the output of the statistical model as additional features. 
Instead, the neural network model can dynamically change its topology, 
following the statistical model's interpretation of the data. In this way, the 
neural network can utilize the graph/tree patterns recognized by the 
statistical model more naturally. We will also build a machine learning toolkit 
that is flexible enough to facilitate the implementations of our paradigms 
where existing toolkits have trouble dealing with.


Date:			Monday, 16 November 2020

Time:                  	2:00pm - 4:00pm

Zoom Meeting: 
https://hkust.zoom.us/j/91228152449?pwd=QUZIMG9VQ2hDRWpPWWYvcllrdkpWQT09

Committee Members:	Prof. Dekai Wu (Supervisor)
  			Prof. Mordecai Golin (Chairperson)
 			Prof. Dit-Yan Yeung
 			Prof. Nevin Zhang


**** ALL are Welcome ****