Our Projects
Oct 2022 - May 2023
MalwareAI
Using Dynamic Reports extracted from Cuckoo Sandbox. we constructed a novel feature vector to classify Malware. Our approach achieved a result of 97.42% accuracy for malware classification, beating previous approaches. Completed under the excellent mentorship of Prof Tram Truong-Huu.
Jun 2023
MALPRED
Using the Kaggle Windows Malware Prediction Dataset, we build a Meta-Model that was tuned with the Tree-Structured Parzen Estimator to predict whether a Windows Machine was likely to get infected by Malware based on it's statistics. Our approach achieved a 73.24% AUC, beating existing approaches.
Aug 2022 -
A Classroom Structure is all you need for Noisy Student Training
We introduce a classroom structure with data splits to improve Noisy Student Training methods.
Jul - Sep 2022
Cura
An AI-powered Social Network for Caregivers to talk. Submitted to Splash Awards 2022 and MindfulHacks 2022.
Nov 2021 - Jan 2022
EmbodiedAI
We use Semantic Segmentation models to segment images of HDB interiors and exteriors. Done as part of the SUTD Research Mentorship Programme 2021.
Jan 2023
Rosebud
Talking to people to network and develop your personal brand is hard. Whether it is at a tech function, forced corporate outing, All adults want to talk about are sports, mortgages and the internal politics of countries they will never set foot in themself, So we made an application that takes the pain out of soPhisticAtIoN, that will listen in on conversations, pick out keywords, and suggest trivia to allow you to speak on the level of adults. Submitted to Hack&Roll 2023.
Oct 2022
FilTeX
filTeX is a content moderation system that comes in two similar but very distinct parts, a chrome extension, and an online virtual chat room. In both platforms, filTeX uses a combination of Computer Vision and Natural Language Processing related models in a multi-modal format to detect a variety of different types of inappropriate content, from general profanities, to inappropriate ASCII art, to simply too mature sentences. Such content is marked via a spoiler by the AI, allowing us to be protected from the text unless we are ultimately willing to do so. Submitted to NTU MLDA Deep Learning Week Hackathon 2022.
Aug - Oct 2019
Netflix Stock Prediction
We have predicted stocks by training neural networks on data from Yahoo Finance. This was done using the Machine Learning Models: LSTM, GRU, CNN and FFNN. First, We found the most optimal architecture for the LSTM model. Based on this, we concluded that a LSTM model yields the most accurate results. In the end we tested against Apple, NVIDIA and DBS to see if it recognised their patterns as well. In the end used it to predict day high and low and finally find the day high and low predictions for the next 10 days (21 September - 1 October 2019).
Jan - Mar 2022
UVAI Prediction
In this project, we aim to train a model to predict the UVAI readings, based on the ambient gaseous concentrations of specific gases in the atmosphere, including Methane (CH4), Sulphur Dioxide (SO2), Ozone (O3), Nitrogen Dioxide (NO2), Formaldehyde (HCHO) and Carbon Monoxide (CO). We conclude that the best performing model is the RBF Kernel SVR. We note that gas column densities and locations may not be an optimal set of predictor variables for predicting UVAI.