Qirui He

Hello World👋🏼 ! I’m Qirui. Welcome to my personal portfolio! Here shows what I’m interested in, working on and learning about.

About Me

🎓Studying Biomedical Informatics @ National University of Singapore: Master’s.

🎓Studied Data Science @ Beijing Normal University-Hong Kong Baptist University United International College(UIC): Bachelors (June 2022).

👨🏻‍💻Interested in Machine Learning, AI and Computer Vision.

🗣Speaks Chinese, English and Cantonese.

📍Born in China, currently in Singapore.

📧qirui_he@u.nus.edu

📋

Resume

Readings

Here I’d like to share some of the papers I’m interested in with my literature review.

📕

Readings

📕

You Only Look Once: Unified, Real-Time Object Detection

Essay

📗

Factorization Machines & Field-aware Factorization Machines

Essay

Projects

Here I post some of the academic projects I’ve done during my undergraduate&postgraduate study. You can find the details and codes of the projects in my GitHub here.

🔬

Academic Projects

🚘

Drivable Area Detection

GitHub:

Summary: Drivable area detection: Using both of traditional object detection method(handcrafted features + machine learning techniques) and deep learning based object detection method to implement the lane detection and car detection respectively. Then combine the lane detection part and car detection part into a complete system for autonomous driving.

Machine Learning

Object Detection

Instance Segmentation

👔

Recommender Systems on H&M Fashion Dataset

GitHub:

Summary: Applied EDA on the H&M fashion dataset to gain insights into the data and implemented Popularity Recommender System, Cosine Recommender System and Pearson Recommender System based on Turi Create. The Popularity Recommender System recommends the top-2 popular items among all items to users while Cosine and Pearson Recommender System recommends the top-12 items that are most correlated to each user’s previous purchases based on the collaborative filtering.

Recommender System

Collaborative Filtering

🫁

Chest X-Ray Images Classification

GitHub:

Summary: Use three different deep convolutional neural network models to deal with the classification of Chest X-image classification for biomedical treatment. The Chest X-image dataset are divided into Bacterial pneumonia, Virus pneumonia, COVID-19 and several other categories. Applied data augmentation, data enhancement and transfer learning methods to the training dataset during the multiple classification problem. The final accuracy of the InceptionResNetV2 achieve an accuracy of 100% in binary classification problem and an accuracy of 97.45% in multiple classification problem.

ICU Intubation Prediction

GitHub:

Summary: Based on the exploratory data analysis on the pre-extracted MIMIC-IV dataset, applied SOMTE as the oversampling method to deal with the imbalanced data. Then implemented Logistic Regression, Decision Tree and Random Forest classifier for the prediction. For each model, applied Genetic Algorithm for feature selection separately and used grid search to finetune the parameters of each model. The Random Forest classifier achieve the highest Area Under ROC of 0.76 with an accuracy of 0.96 on predicting the patient’s chance of intubation in ICU.

Machine Learning

Feature Selection

EDA

Technical Skills

Languages

Frameworks&Tools

📈TensorFlow 🔥PyTorch 🐼Pandas 🟦NumPy 🟠Scikit-learn 🐳Docker 🐧Ubuntu

🅚Keras 🌍Matplotlib 🧮Math 🌊Seaborn 📊Tableau ⚡️Spark 🧱Django