cv

Undergraduate Teaching Assistant

Halıcıoğlu Data Science Institute, University of California, San Diego

Develop programming assignments, prepare starter code, and write unit tests to grade submissions on Gradescope. Helped 500+ student by holding regular office hours and answering questions on EdStem.

• Teaching
• Gradescope
• Assignment Development

Note-Taking Volunteer

Office for Students with Disabilities, University of California, San Diego

Provide legible lecture notes to students with disabilities.

• Across five STEM courses

Bachelor of Science

University of California, San Diego

Data Science, Mathematics-Computer Science

• Theory of Relativity

Summer Visiting Student

University of California, Berkeley

Statistics, Computer Science

• Introduction to Artificial Intelligence
• Concepts of Statistics

ACM AI Quarterly Project

Conducted EDA on an imbalanced kaggle dataset and built an automated pipeline for model training. Applied transfer learning with pretrained BERT (Bidirectional Encoder Representations from Transformers) model. The model classifies input text corpus into 1 of 16 MBTI personality types with ≈ 95% validation accuracy. Deployed a Streamlit App that allows users to type in words and obtain classification results during showcase.

• BERT Model

Semantic Segmentation with PASCAL VOC 2007

Performed data augmentation and designed an class weight heuristic to deal with small imbalanced dataset issue. Implemented Fully Convolutional Network (FCN) baseline model and UNet architecture. Used IoU value, along with pixel accuracy, to evaluate model performances due to dominating background pixels. Leveraged the FCN ResNet-101 pretrained model to improve IoU from 0.0566 of the FCN baseline to 0.330.

• ConvNet
• UNet
• FCN ResNet-101

Game Recommendation with Steam Dataset

Implemented and trained Bayesian Personalized Ranking Model, and ensembled with popularity-based recommendation method to predict if a person would play a given game; ranked top 2% on the course leaderboard. Trained a Latent Factor Model with bias terms only to predict a person’s play time on a given game.

• Recommender System

Little Pac-Man

Formulated Minimax, Alpha-beta Pruning, Expectimax adversarial search methods for Pacman to survive. Implemented Bayes Nets and Hidden Markov Models for inference tasks, such as predicting ghost positions.

• Adversarial Game

Pattern Analysis on EEG for Enhanced Depression Diagnosis

Explore patterns between brain EEG data and severity of depression with OpenNeuro dataset. Conduced EDA to analysis noisy EEG data and identify main features through PCA and ICA. Potentially incorporate ML/DL models to extract features through learning to help pattern analysis.

• EEG

Facial Emotion Detection

Explore ConvNet Architectures to build classifiers to recognize facial emotion given image inputs. Selected project lead; facilitated group meetings, searched for resources, and communicated with team members.

• Computer Vision