Newt KHN Nguyen 阮金喙南

Newt KHN Nguyen 阮金喙南

he/him

HKUST

Publications and Projects

CIPM: Convolutional Imaging Prediction Model for Cryptocurrency Price Prediction

Published in For partial fulfilments of COMP 4211 Requirements, 2024

Time series prediction has long been a key field of research across numerous fields, including finance, economics, and computer science. As the number of machine learning tools available for assisting decision-makers in making predictions continues to grow, there is growing interest in leveraging these advanced tools to enhance timely decision-making. In this report, we present a novel application of machine learning aimed at predicting price trends in cryptocurrencies. Cryptocurrency has recently been gaining traction and establishing itself as a significant asset class. We utilize Convolutional Neural Networks (CNNs) because they mimic the way traders and human make decisions by analyzing trends in images. These trends are particularly notable in cryptocurrencies due to their highly volatile nature. We train and validate our model on 9 different coins and Bitcoin (BTC), then test our model on the 9 coins, primarily by predicting whether their price will go up or down in the next hour. We will compare the performance of our CNN with a basic Long Short-Term Memory (LSTM) model trained on time series data.

Recommended citation: Darren Chua, Newt Nguyen (2024). "CIPM: Convolutional Imaging Predictional Model for Cryptocurrency Price Prediction".

Replication of Resilient K-Clustering (KDD, 2024)

Published in For partial fulfilments of COMP 5331 Requirements, 2024

Clustering is a crucial problem in the realm of databases. It plays an important role in data analysis and pattern recognition. One of the widely recognized algorithms in this domain is K-clustering. However, K-clustering is sensitive to various factors, including outliers, centroids initialization and input perturbations. These characteristics increase the computational cost in attaining the globally optimal solution since multiple runs of the algorithm are often required to achieve more stable and reliable clustering outcomes. In light of these challenges, there has been a growing interest in exploring approaches that can enhance the stability of traditional K-clustering algorithms. The resilient K-clustering algorithm, proposed in Ahmadian’s paper, represents a recent advancement that demonstrates improvement in the resilience of clustering algorithms towards input perturbations. As there is no implementation of the paper before, for this project, we will implement the paper comprehensively.

Recommended citation: Yoanna Lo, Jason Li, Newt Nguyen, Kelvin Tam, Dennis Tsang (2024). "Replication of Resilient K-Clustering (KDD, 2024)".