WHO I AM

I am a Deep Learning Researcher and Data Scientist interested in impactful R&D for complex problems in Deep Learning. Currently a Graduate student at Stevens Institute of Technology finishing a M.S. in Computer Engineering with Graduate Certificates in Artificial Intelligence in Engineering Design and Cybersecurity. I'm also a part-time engineer at MITRE performing research in new domains along with my independent research projects.

From an early age I fell in love with computers and have been building, tinkering, and programming them since then. A crippling injury to a close family member in my early adolescence inspired me to persue robotics and their applications for prosthesis. I found Stevens which let me persue my interests which lead me to focus more on the intelligence side of robotics before uncovering Machine Learning at their (former) High-Performance Computing center. This evolved into research seminars on the burgeoning field of Deep Learning where I discovered a true passion for it.

I held the leadership position of the President of the Stevens Branch of IEEE where I conducted many informative hands-on workshops including some on Deep Learning. I've also been involved in research concerning Neural Networks and Natural Language Processing, focused on extracting text and sentiment analysis on news sources. My Senior Design (Capstone) project involved using Neural Networks to detect cell tower spoofing and Man-In-The-Middle attacks using only signal attributes from a user's phone. Currently, I am currently working on a Master's Thesis concerning the implemntation of Capsule Neural Networks in a Generative Adversarial Network (GAN) architecture for Semi-Supervised learning.

For more information about my research and projects check out the rest of the page. I am a very adaptable, self-motivated learner who wants to use my skill set to help people. I am also an avid traveler and an amateur photographer. I'm finishing my Master's of Science in Computer Engineering at Stevens in December 2018 and am looking for a position in Deep Learning Research starting February 2019. Feel free to reach out to me through LinkedIn.

x

Generative Capsule Neural Networks

Creating a Semi-Supervised Capsule Neural Network image Generator

My ongoing Masters' thesis where I am studying the viability of creating a Semi-Supervised Capsule Neural Network (CapsNet) image generator. Generating from the output of a CapsNet was proposed in the original paper 'Dynamic Routing Between Capsules' In order to accomplish this a CapsNet was trained for classification and the output of a single class in the DigitCaps layer was input into a simple Dense Decoder and trained to generate an image from this output. This method is not Semi-Supervised which is why a Generative Adversarial Network (GAN) architecture may be able to accomplish Semi-Supervised Learning. CapsNEt will be used both as the generator and critic/discriminator positions of this GAN model. Check out the Github Repo for an active look into how it works and sample images.

Current Status: Initial tests have confirmed that a CapsNet can be used as a generator for the MNIST handwritten digits data set with another CapsNet used as a discriminator. There is still some improvements that can be made to help stabalize training, but other testing has shown that a slightly modified CapsNet generator can also generate images from the CIFAR-10 data set if trained properly.

Project Repository
x

Live Memetic Detection

Classifying images live on Social Media feeds

I trained a R-CNN with Tensorflow Object Detection API capable of detecting images of “memes”, or viral photos with a specific message or parody, on a live screen capture of a social media feed. I was able to distinguish memetic images from normal images that contained similar characteristics, such as advertisements. I wrote scripts to scrape and collect images used for training from image search engines and used Python and Bash to streamline labeling and the data sanitation process.

Current Status: Project was successful and our live demo was able to robustly distiguish between the Geico gecko, Kermit the Frog, and pictures of Pepe the Frog on different live feeds. The Github repository below contains a proof of concept CNN for distinguishing between pictures of the "Doge" meme and regular user taken photos of dogs.

Project Repository
x

Iceberg and Ship Detection Class Project

Classifying ships and icebergs from space

Final class project focused on distinguishing between ships and icebergs on geo-spatial data labeled by human experts taken at different angles from satellites. I created a CNN based on the VGG architecture to correctly classify between the two on a large validation data set with a high validation accuracy. This dataset was originally part of a $50,000 Kaggle challenge and the results of the class project would have been competetive.

Current Status: Completed the project which can be found in the Github repository link below. The data can be found on Kaggle here.

Project Repository
x

Neural Network Security

Leveraging Neural Networks to detect man-in-the-middle attacks on cell phone connections

Senior Design project through Stevens Institute of Technology whose goal is to create a revolutionary system that will monitor and detect suspicious network activity, such as network spoofing and man-in-the-middle attacks using advanced machine learning algorithms. Our system will be able to monitor GSM packets in order to provide comprehensive user protection. Users can access data from our machine learning algorithms using our front end portal.

Current Status: We currently have an app that streams mobile data such as signal strength and baystation IDs from a user's phone. The app is being flushed out as we update the interface to give users more functionality and data about their mobile connection. Using real world scenarios such as signal type, user distance to a cell tower, and test cases of man-in-the-middle attacks we simulated 30,000 scenarios, half of which were normal cell phone use and the other half were results from a man-in-the-middle attack. Using these 30,000 cases we were able to train a 3-Layer Neural Network that provides a success rate of 99.999% on an independent validation sample of 3,000,000 scenarios. Even better, the misclasifications by the neural network only result in false positives and does not miss any man-in-the-middle attack scenarios. A real-time demo will be presented on May 3rd at the Stevens Senior Innovation Expo at Stevens Institute of Technology.

Follow this project
x

Sentiment Analysis

Using python to extract data and analyze sentiment from news articles

Independent project advised by a Stevens Professor. Leveraging the variety of text processing and neural network libraries available in Python, I scrape trustworthy news feeds and extract pure text from the pages. The text is then processed using Natural Language processing to determine a positive or negative sentiment. This information is then stored into a data set that can be processed by a neural network to correlate the sentiment with other available data sets, such as stocks.

Status: Automated text extraction and sentiment analysis are complete and fully functional. Currently, I'm formating data so it can be digested by a neural network.

PORTFOLIO

Generative Capsule Neural Networks

Creating a GAN using only CapsNets

Live Memetic Detection

Classifying images live on Social Media feeds

Iceberg and Ship Detection Class Project

Classifying ships and icebergs from space

Neural Network Security

Securing your phone data

Sentiment Analysis and the News

Making sense of the news
SKILLSET

Skills

To see samples of my code visit my Github.

Programming

Python (2.7 +3.X)
Java
C++
Matlab
CUDA
Verilog
VHDL
Bash
HTML
CSS
JavaScript
jQuery
Basic

APIs

Keras
Tensorflow
Theano
NLTK
Jupyter
boilerpipe
NumPy
scikit-learn
Spark
Matplolib
Pandas

Operating Systems

Windows 7
Windows 8.1
Windows 10
Ubuntu
UberStudent
Kali
Arch
OS X Yosemite

Software

Docker
Git
Android OS
Sublime Text
GNU nano
VIM
Geany
Firebug
Eclipse
JetBrains Suite
Parature
SolidWorks
MS Word
MS Excel
MS PowerPoint
Adobe Photoshop 6
Adobe Flash
Dreamweaver

Interests

General Artificial Intelligence
Machine Learning Applications for Industry Sized Problems
Biological and Developmental Psychology
Cybersecurity
Astronomy
Nature Conservation
Tech Policy

Activities

IEEE
Music
Hiking
Running
Travel
Photography
Film

Download Resume
PHOTOGRAPHY