Explore Semi-supervised

Learning on Hypergraphs

FYP16005 Final Year Project

Learn more

OUR TEAM


Hubert Chan

Project Supervisor

Email: hubert@cs.hku.hk

Zhang Chenzi

Supervisor Helper

Email: czzhang@cs.hku.hk

Chen Jiali(Charlie)

Project Member

Email: jchen@cs.hku.hk

Zhang Ying(April)

Project Member

Email: yzhang16@cs.hku.hk

INTRODUCTION


This project aims to study and explore a hypergraph based semi-supervised machine learning algorithm. In this project, the studied algorithm has been implemented. The original binary classification algorithm has been generalised to multi-class classification version. Several activation funcitons have been applied seeking higher accuracy and new regularization function has been substituted for faster convergence and new understanding of the learning process. Parallel computing has been adopted. Multiple experiments have been conducted to examine our hypotheses.

DOCUMENTATION

Click on the preview to download files.


Project documents

img

Project Plan

img

Interim Report

img

Final Report(Charlie)

img

Final Report(April)



References

img

Gaussian field and harmonic function

img

Learning with hypergraph

img

Total variation on hypergraph

img

Learning with Local and Global Consistency

PROJECT STEPS




Research

Research on the sub-gradient approach and total variation approach, understand the mathematical theories

Implementation

Implement two approaches and perform testing with data from the machine learning repository. (link)

Analysis

Analyze the deficiency or the limitations of the current methods, plan for further improvements.

Interim Report

Deliverables of interim report
Interim presentation rehearsal.

New Method

Deliver mathematically our own method according to former analysis. Perform implementation and analyze its accuracy as well as time and space complexity

Final Report

Deliverables of final report.
Final presentation rehearsal.

Project Poster

img

Click on the image to see HQ version.