Singapore, Singapore
Hello, I am Stacy, a fresh graduate from NUS with a major in Data Science & Analytics. I am passionate about leveraging AI to solve real world problems and wish to pursue a career in AI. In particular, I enjoy studying the mathematical foundations that underpin AI algorithms to understand not just how they work, but why they work.
- Conducted Mathematics and Science lessons for 8 primary and secondary school kids weekly and provided concise notes and tailored worksheets, resulting in at least one grade improvement for each student
- Cleaned and Preprocessed a tabular PCOS dataset consisting of clinical and diagnostic features - Trained traditional (Random Forest Classifier, Logistic Regression) and deep learning classification models (Multi-Layer Perceptron) on the dataset in Python - Explored transformer-based models such as TabTransformer and implemented a novel transformer-based model adapted from TabTransformer's architecture using Keras
- Cleaned and Preprocessed an unlabeled dataset containing approximately 4.5 million Reddit comments from 2020 to 2023 across three Singapore subreddits - Used an open-source labelled Jigsaw dataset containing approximately 220,000 comments from Wikipedia talk pages to evaluate 3 pre-trained BERT-based classification models (Toxic-BERT, RoBERTa and Dehate-BERT) - Labelled the Reddit comments as toxic or non-toxic using the RoBERTa model which has the highest roc auc score and assessed the changes in the level of toxicity in Reddit over the years using the labelled Reddit dataset - Developed an unsupervised BERTopic model to identify the key drivers for this trend and gave recommendations on how to lower the toxicity level
Melanoma Prediction Using Multi-Modal Data - Developed classification algorithms to predict patients with Melanoma, based on patients' skin lesion images and demographic data in Python, potentially improving dermatologists' diagnostic accuracy - Preprocessed 16437 images by cropping using the segmented images and converting to arrays - Extracted features from images using Keras' ResNet-50 and constructed an xgboost classifier, achieving 0.9 roc auc score - Built a multi-modality attention network with cross-attention blocks with Pytorch and used vision transformers pretrained by masked autoencoders for image encoding, obtaining an roc auc score of 0.84