Selected Projects
K-pop face generator
A generative model to create faces that look like K-pop idols. I experimented using both a VAE and GAN to try and model the prior distribution. Overall I was able to create some decent looking images although I wouldn't say they are super convincing. To improve this I would probably train on colab with better GPUs so that I can increase the model size and the resolution of the images.
DABNet
A reinforcement learning model to play Dota Underlords. I used OpenCV to do a lot of the visual processing, and then implemented DQN algorithm to learn the which actions to take given a certain board state. Overall after a few thousand epochs it was able to win the first few rounds of a match against humans.
Meowtalk
A web app which uses ML to provide different playlists based on a persons emotions. For this I implemented a sentiment analysis LSTM model which utilizes GloVE embeddings for the embedding layer. This was able to have a decently high accuracy on the dataset, and worked well for our hackathon demo.
ML Algorithm Implementations
Random ML algorithm Implementations. I usually tested the algorithms on the Iris dataset. Most of these are just algorithms I learned in classes.
Fake News Validator
A research project to prevent the spread of Fake News. My partner and I developed an architecture to both detect fake news, and suggest articles which cross-validate the information proposed in the Fake News. We did this using hierarchical attention, and transformer models for suggestion and detection.
Job Post Generator
An RNN language model to create job posts. This was created for a hackathon. I had to develop both the data gathering script, and the model training. It was able to generate a few realistic prompts, but definitely needed more data and training time.