Brachial Plexus Segmentation from Ultrasound Images using Res-U-Net
As part of a Kaggle Competition used deep architectures to generate a mask of the brachial plexus from ultrasound images. The best model from my experiments was in the top 8% of the competition consisting of 923 participants. A dump of code experiments can be found: here.
Mesh Networking Android App
A prototype, hackathon-esque mesh networking chat application for the Android using Wi-Fi Direct. Android code can be found here. Laptop client code can be found here. And a report on it can be found here.
Autonomous Robot for CTF Competition
Programmed an autonomous robot that won 1st place in the Design Principles and Methods class competition. Code on my Github. Here is a video of it (thanks to one of my teammates) and poster can be found here.
Halma Playing AI
Programmed an AI that plays a modified team version of the board game Halma. Quick write up about it here (extremely quick so don't judge), placed in the 98th percentile of a class wide competition with a relatively simple heuristic based search. Also on my Github.
A networked multiplayer 5 Card Stud poker game written entirely in Java (using Swing of all things!). Also on my Github.
Canadian Open Data Experience
Built a website for seeing and reporting food warnings from restaurants or products. Parses the data provided by the Canadian Health and Safety service, uses MongoDB, Node.js, HTML. Also on my Github.
McGill: Code Jam
Built an automatic securities trading machine in Java for a Code Jam sponsored by Morgan Stanley. Also on my Github.
Fully Convolutional Character-level Dialogue Response Generation
We present the first fully-convolutional character-level dialogue response generation system based on the ByteNet architecture proposed by Google DeepMind. We also modify the system to promote diversity and compare the system against a commonly used LSTM-based architecture with attention. We train both on the Cornell Movie Dialogue corpus and find that the two are comparable based on BLEU score and human judges. However, due to the small size and noisy nature of the Cornell Movie Dialogue corpus, neither the baseline nor the fully convolutional system yield particularly satisfying results. We suggest further work to train the system on a larger corpus and modifying the architecture further to generate more coherent and robust responses.
Download: Report on Initial Findings
Extending H-MAX with End-Stopped Simulations for Biologically Inspired Object Recognition
Here we modify the HMAX object recognition system of Serre et al. with an end-stopped filter to try and improve the accuracy of the model and to further the parallels in the algorithm to the actual processing of the visual cortex for learning object representations.
Download: Full Paper
Multilingual Sentiment Analysis of Twitter Posts
Using modified Deep Learning techniques presented by Socher et al., show the accuracy of a language-independent sentiment analysis algorithm. Training on a combined and separated sets, the learner determines a grammatical and sentiment model for each, then its performance is determined using k-fold validation.
Russia’s Transfer From Communism to Capitalism: A Poor Economic Model and Its Aftermath
This was me delving into the world of anthropology freshman year. On rereading this, it could definitely use some clean-up linguistically, but there's quite a bit of research poured into this paper, so I wanted to share it.