City Skyline

Machine Learning Projects

MSDT: Maksed Language Model Scoring Defense in Text Domain

Proposed a novel improved backdoor defense method in text domain using Maksed Language Mode Scoring metric. "Independent Work Research (COMP4971D)" under the supervision of Professor, Minhao Cheng (HKUST)
[code] [report]

MSDT.png

General Methodology of MSDT

Impact of Adversarial Training on the Robustness of Deep Neural Networks
 
Paper "Impact of Adversarial Training on the Robustness of Deep Neural Networks" has been accepted by the 2022 International Conference on Modeling, Simulation and Computing Science (MSCS 2022).

Experimented the effectiveness of various methods of  adversarial training on improving the robustness of neural networks against classifying perturbed histopathological images. 

[code]

Histopathological Image Classification
 
Metastatic cancer diagnosis based on histopathological image using Convolutional Neural Network and modified Resnet-18.
[code] [slides] [report]

Seq2Seq Neural Machine Translation (Keras)
 
Neural machine translation using Tensorflow framework. Translation processed from English to French
[code] [slides

Adversarial Attack and Defense Presentation Project
 
COMP4211 Machine Learning course Final Presentation Project on the topic of "Adversarial Attack and Defense".  [slides] [video]

 

Papers reviewed: 

  • “Explaining and Harnessing Adversarial Examples"

  • “Is Bert Really Robust? A Strong Baseline for Natural Language Attack on Text Classification and Entailment”