CS 224N final project at Stanford. Preprocessed the AQUA-RAT dataset and extracted mathematical equations from multiple-choice explanations with regular expressions, then implemented a Transformer network and a bidirectional GRU-LSTM with attention to generate equations that solve algebraic word problems.
"}},{"node":{"frontmatter":{"title":"AI Agent for the Atari Game Phoenix","tech":["Deep Q-Learning","MCTS","OpenAI Gym"],"github":null,"external":"/files/CS_221_Poster.pdf"},"html":"CS 221 project at Stanford. Built an AI agent to play Phoenix using a Deep Q-Network, and implemented Monte Carlo Tree Search alongside teammates as a faster alternative approach.
"}},{"node":{"frontmatter":{"title":"Video Interpolation of Human Motion","tech":["3D CNN","Recurrent CNN","Encoder-Decoder"],"github":null,"external":"https://cs230.stanford.edu/projects_spring_2018/reports/8290324.pdf"},"html":"CS 230 project at Stanford. Built an encoder-decoder baseline to interpolate intermediate video frames given start and end frames, then designed 3D convolutional and recurrent CNN architectures and tuned hyperparameters to improve interpolated frame quality.
"}},{"node":{"frontmatter":{"title":"Sensor Network Localization","tech":["Convex Optimization","SDP","ADMM","MATLAB"],"github":null,"external":null},"html":"CME 307 project at Stanford. Tackled the Sensor Network Localization problem using second-order cone, semidefinite dual and least-squares methods, and applied ADMM with permutations to speed up solving 3D SNL problems with 10 sensors.
"}},{"node":{"frontmatter":{"title":"Fake News Stance Detection","tech":["scikit-learn","SVM","Naive Bayes","Feature Engineering"],"github":null,"external":"http://cs229.stanford.edu/proj2017/final-reports/5244160.pdf"},"html":"CS 229 project at Stanford. Extracted n-gram, sentiment, polarity and cosine-similarity features from 75,385 news articles, then compared multinomial Naive Bayes, SVM, softmax and a multi-layer neural network to predict whether a headline agrees with, disagrees with, discusses, or is unrelated to its body.
"}},{"node":{"frontmatter":{"title":"Gaussian Mixture Models for Digit Clustering","tech":["MATLAB","PCA","Expectation Maximization"],"github":null,"external":null},"html":"Math 156 project at UCLA. Ran PCA on MNIST to reduce dimensionality, implemented the Expectation Maximization algorithm for Gaussian Mixture Models from scratch in MATLAB on 14,000 training samples, and benchmarked the clustering against MATLAB's k-means.
"}}]}}}