Deep Learning- Advanced Course

First year at KTH

Siamese Masked Autoencoder: Paper Reproduction, Link

We have used the PyTorch framework to reproduce a semi-supervised multi-object segmentation model, which extends the Masked Autoencoder. The authors have incorporated a Siamese network into the Masked Autoencoder, enabling it to outperform some state-of-the-art (SOTA) models like VideoMAE and Dino.

My contribution:

• Model Building and Validation: Responsible for constructing, evaluating, and visualizing the results of our models to ensure accuracy and efficiency.

• Report Writing: Tasked with compiling comprehensive project documentation and results analysis.
• Training and Management: Managed the training of models on Google Cloud Platform (GCP) and maintained our project’s codebase on GitHub.

Big Data Analytics

First year at NYCU

DL application-Food Classification using Tensorflow and Anvil web APP, Link

We used deep learning and ANVIL's product to create an interactive interface. 

My contribution: 

• Construct the deep learning model for the app using Transfer Learning techniques with EfficientNetV2S as the base model.
• Developed a model, the Domain-Selection-Model, to select between two models trained on distinct datasets for making predictions. 

Deep Learning

First year at NYCU

Deep learning application-Self-driving Robot simulation using PyTorch, Link

We built an image recognition deep learning model to do the self-driving car simulation.

My contribution:

• Data augmentation and data pre-processing.
• Construct the deep learning model for the app using Transfer Learning techniques with ResNet50 as the base model.

Machine Learning

Senior year at NTHU

Deposit Subscription Prediction using R, Link

We implement several statistical-based machine learning methods to predict whether the customers will subscribe to the deposit service or not. 

My contribution: 

• LDA, QDA, KNN, and Naive Bayes, four statistical-based machine learning methods, to make predictions using R.