View Proposal

---

Proposer

John See

Title

Self-Supervised Learning for Analysis of Facial Micro-Expressions

Goal

An algorithm/model for performing multiple tasks on facial micro-expresisons (i.e. spotting, recognition), with further insights into its feasibility and drawbacks

Description

The self-supervision learning (SSL) paradigm -- which allows for models to be trained on a task using the data itself to generate supervisory signals rather than relying on externally-provided labels-- has opened up new possibilities of having one generalised model to cater for a range of downstream tasks. Meanwhile, computational analysis of facial micro-expressions has been gaining attention among the affective computing community in the last decade due to the proliferation of new datasets. The field comprises several associated tasks, such as spotting and recognition, which may qualify as potential downstream tasks using a specially crafted SSL method. This project aims to investigate the feasibility of SSL models in several micro-expression analysis tasks and to discover insights that will be valuable to the community. This is a research-centric project.

Resources

GPU compute (access to MACS Malaysia server OR supervisor's GPU workstation)

Background

Strong competency in algorithms and programming, especially Python; Some familiarity with machine learning or deep learning techniques would be an added advantage.

Url

Difficulty Level

High

Ethical Approval

None

Number Of Students

0

Supervisor

John See

Keywords

machine learning, self-supervised learning, micro-expressions, affective computing

Degrees

Bachelor of Science in Computing Science