- Focused on building an multi-modal, interpretable and editable recommendation system that leverages Multi-modaland Large Language Models for making predictions with production level scalability and speed.- Interacted with different production-level datasets which the method will be applied for real-world applications.

Project 1: COVID-19 Patient Risk Stratification under Cost Constraints- A fast, easy-to-use and explainable patient risk score prediction model that aided doctors to calculate patients’ need for hospitalization and ventilation or chance of death based on their electronic health records and costs of lab tests- Implemented data imputation, feature engineering and classification methods on incomplete and imbalanced data- Compiled and composed related works, implementation details and… Show more Project 1: COVID-19 Patient Risk Stratification under Cost Constraints- A fast, easy-to-use and explainable patient risk score prediction model that aided doctors to calculate patients’ need for hospitalization and ventilation or chance of death based on their electronic health records and costs of lab tests- Implemented data imputation, feature engineering and classification methods on incomplete and imbalanced data- Compiled and composed related works, implementation details and results for three medical journal publicationsProject 2: Failure Incipient for Power Transformer- An end-to-end pipeline for classifying health status of electric power transformers by modeling voltages and microPMU data using stochastic processes such as Chinese Restaurant Process- Designed data processing pipelines that convert raw multi-modal data from into a unified ready-to-use format- Composed and proofread related works, experiment details and implementation details for an accepted publicationProject 3: Implementation of FastCAM to Captum- Adding FastCAM, an saliency map method, to Facebook’s open source deep learning feature attribution library Captum - Implemented the main algorithm and test cases that are ready for open source contribution standards- Surveyed and presented an overview of modern explainable AI methods for the machine learning group at LLNL Show less

- Designed skeleton and graded all of students’ homework and final presentations- Provided help to students through online forum by answering questions related to lecture or logistics daily

- Research Focus: Interpretable Deep Learning, Sparse and Low-rank Methods, and Computer Vision.- Two publications on Maximal Coding Rate Reduction (See Google Scholar for paper)

Project: Subspace Locally Competitive Algorithm- Published as second author (check Google Scholar for link)- Analyzed plots and contributed to insights in weekly research meetings- Contributed to mentor’s open source deep learning repository on GitHub- Compared code in open source library Tensorflow and verified outputs with different implementations

Project: Understanding Facial Recognition Models through Local Feature Attacks- Deployed an end-to-end AI pipeline and systems for real-life applications- Created an API for testing model robustness to diverse input images and adversarial examples- Presented to other researchers on new adversarial attacks and security flaws in deep learningProject: Model Optimization by Neural Network Pruning- Experimented on different pruning methods and optimized deep learning model… Show more Project: Understanding Facial Recognition Models through Local Feature Attacks- Deployed an end-to-end AI pipeline and systems for real-life applications- Created an API for testing model robustness to diverse input images and adversarial examples- Presented to other researchers on new adversarial attacks and security flaws in deep learningProject: Model Optimization by Neural Network Pruning- Experimented on different pruning methods and optimized deep learning model performances- Summarized common pruning and optimization methods into a write-up for other researchers- Presented the analysis and plots of experimental results in monthly department meetings Show less

Project: Neural Network Parser for Resume Recommendation System- Improvised and applied concepts from multiple academic papers to improve recommendations- Processed 100k+ Chinese sentences to speed up data processing by 100-150%- Developed server-pressure tests for product’s runtime and memory performance analysis- Set up server configuration and defined corporate coding style to improve developing workflow

- Taught K1-K12 student mathematics ranging from Basic Arithmetic to Single Variable Calculus- Finished Mathnasium's tutoring training and trained specifically to teach "number sense"- Practiced time management skills and social skills by interacting with younger students

- Provided math tutoring for 30 students each week- Tutored Math level Basic Arithmetic to Linear Algebra- Provided academic advises to educational and career planning