CIFAR-10 on Kaggle with MXNet (Computer Vision)
Independent Project • Oct 2017 — Nov 2017
Hands on different CNN architectures to identify the subject of 60,000 labeled images
- Achieved 95.34% accuracy.
- Earned Top 5 Award in Amazon MXNet Community by implementing AlexNet, VGG, ResNet and DenseNet in MXNet-Gluon, leading to 95.34% ensemble accuracy.
- Reduced training time from days to hours by deploying Amazon AWS GPU instances based on NVIDIA CUDA.
Team Leader & Primary Developer • Feb 2017 — May 2017
To build a system for auto-recognition of the age, gender and OCEAN personality
- Achieved the highest accuracy among twenty teams w.r.t. all 3 targets by ensemble learning 12 models.
- Used image data by transfer learning Google Inception-V3 and implementing 2 CNN models in TensorFlow.
- Used text data by applying Naïve Bayes and TF-idf based SVM classifiers in Python.
- Used users’ likes by implementing KNN, page-user-page, SVD based LR, and a Perceptron NN in Keras.
Smart Light System (iOS App + AWS-IoT)
Individual Project • Apr 2017 — May 2017
First attempt to intelligent homes. Not only can it auto-control lights, but also managed remotely
- Accomplished a remote light control system by developing a client-side mobile app in Swift and employing AWS Congnito and AWS IoT as server-side.
- Employed PIR motion sensors to automatically control lights under auto mode.
Individual Project • Jan 2015 — Jan 2016
To develop a mobile based app to help both early-stage dementia patients and their carers
- Top-rated final year project of University of Liverpool in 2016, and is kept in campus library.
- Accomplished this project by implementing an iOS app in Objective-C with MySQL and PHP based backend.
- Allowed early-stage dementia patients live at home with daily issue reminder, built-in map navigator, and an instant message system. Enabled caregivers to track patients’ daily issue progresses and real-time locations.
- Ensured robust services and database consistency under poor network connectivity by employing local SQLite.
Cloud Campus (Social Networking Website)
Founder & Primary Developer • Jan 2015 — Jun 2015
To promote organizations’ developments and enhance students’ communications
- Used HTML/CSS, PHP, JavaScript and MySQL to implement the website.
- Students can not only connect friends and share their moments as traditional social webs, but also share their plans and events in real-time with friends.