Judy Stephen

Leaf Classification

Machine Learning (Neural Network and Logistic Regression)

Classification of plant life has been historically arduous. In a Kaggle playground, members were asked to make classify 99 species of plants based on binary images and extracted features such as shape, margin and texture.

My model ranked 29th place out of 1200 submissions, placing it at top 3% with a score of 0.00581 (closer to 0 the better the score). A convolutional neural network with 2 hidden layers was used to achieve the results. Originally, I had used an optimized logistic regression classifer which ranked 58th on the leaderboard to be in the top 5%.

Lexer and Parser

Programming Languages, Compilers

Programming languages are interesing, fun and some what of a mystery in the beginning. I wanted to learn more about languages and understand the entire process of going from a high-level language which is easily readable by humans to something that will eventually be translated to 1s and 0s. I learnt about compilers and wrote my own interpreter for the BASIC programming language using Python. Then I progressed to write a lexer to tokenize basic C++ constructs and a parser to create a abstract syntax tree.

Pet Robot

Speech Recognition & Computer Vision

As a final project to ECE 5725 Embedded Operating Systesm, my partner and I wanted to create a project that would resonate with regular consumers while incorporating speech recognition and computer vision. We decided to make a pet robot that utilized both of these techniques to create a user friendly product.

Our robot was capable of listening for user commands, as well as replying to specific commands, such as saying today's date. The robot was also able to play a modified version of fetch. When users tell the robot to fetch, the robot will search for a blue object and move towards it. After finding the object, it signals to the user that it found the blue object and then will search to find a green object. Once it has found both the blue and green objects it will be done playing ball and wait for a user command.

Sports Reward Android App

Hackathon

As a part of Big Red//Hacks 2014, my team and I built an Android App that allowed fans to get rewards points everytime they went to a sporting event. When the users go to sporting events, the app will automatically let them in using our custom Bluetooth module. They will also be able to explore nearby restaurants, bars and hotels to the stadium using our app. We utilized Factual and Priceline APIs to provide these relevant information.

We won both Oracle's Best Hardware Hackathon and Yext's best use of location services awards.

World Mood with Twitter

Arduino

I wanted to explore Twitter API so I built a world mood analyzer. The Python program will use the Twitter API to gather Tweets and use a Naive-Bayes classifier to get an average mood of the world based based on the Tweets. The mood will then be displayed using an LED controller with an Arduino. For example an excited mood would make the LED glow red, while sad mood will make it glow blue.