My Résumé

Resume

Publish Date: 2022-09-06

Dailun Li

Room 404, 1910 Oxford St, Berkeley, CA, 94704 (510) 316-4347 | lidl200147@gmail.com

GitHub Repo: github.com/lidl2019

————————————————————————————————————————————————–

University of California, Berkeley, United States Expected Graduation: 2024.6

McGill University, Montreal, Canada 2019.9 - 2023.1

University of Oxford, Oxford, United Kingdom 2021.4 – 2021.7

————————————————————————————————————————————————–

ChromX Health Ltd., Guangzhou, China

Developed two techiques on evaluating mass spectrums on people’s breathing materials to help predicting Lung Cancer (Specifically Squamous Cell Carcinoma and Adenocarcinoma). These methods involve machine learning techiques like PCA, Logistic Regression, MLP that consists of 2216 sample points and over 400 chemicals.
Developed algorithms to support wave detecting, wave alignment, and wave separation to measure the specific magnitude of each chemical on the spectrum.
The two algorithms (one using preprocessed wave magnitudes, the other using data points on mass spectrums) are able to robustly separate malignant cancer patients from healthy people, with a 94.6% F1 score over 628 healthy and 605 patients.
Typical extracted biomarkers including Isoprene and Acetone are proven to be identifiers of lung cancer.

McGill University, Montreal, Quebec

Supervised by Professor Bettina Kemme
Help formalized and implement operations for SQL support in AIDA, using pandas-like APIs. Such implementation will turn transformations in pandas to SQL table operations in order to speed up the process. E.g. A scheduler will detect and estimate the operation time and to decide whether the operation will perform locally, or perform in a remote database.

Beijing Venustech Technology Ltd., Beijing, China

Develop web pages in Vue.js that monitor network stream for abnormal message detection
Develop a Honeypot system that analyzes attempts by malicious users
Use Logistic Regression model to label abnormal messages, which has a 78% accuracy
Store user request logs to MySQL database using sqlalchemy

University of Oxford, Oxford, United Kingdom

Supervised by Professor Maarten De Vos
Collect human sleep EEG signals and analyze the corresponding sleep qualities with RNN
The RNN model reaches an 87.5% accuracy on 5-category classification over 5,600,000 samples

*Report can be found at https://github.com/lidl2019/Automated-EEG-grading, which is going to be published at IEEE Explore in late 2022

————————————————————————————————————————————————–

Explore the effects of datasets on the self-attention mechanism with datasets such as Yelp, Xinhua, etc.
Explore the effects of various types of RNN models (LSTM, GRU) on the self-attention layer
Confirm that the self-attention layer may cast negative influence, which contradicts common knowledge

Implement a CNN classifier that classifies handwritten digits and alphabets in 52x52 images
Experiment with various CNN models including AlexNet, VGG-16, ResNet, etc.
Apply random forest, data augmentation, and data filtering to optimize performance
Reach a 96.5% accuracy, which surpasses manual precision, and stand as the top team in 86 teams

Implement CYK parser for French grammar from scratch
Implement TFIDF vectorizer, the CYK algorithm, parsers that converts CFG into CNF
Define detailed French grammar in CFG (Context-free Grammar)