CollabCORE

Supervisors:

• Prof. Jörg Kienzle
  
  (McGill University, Canada)
• Prof. Omar Alam
  
  (Trent University, Canada)

TouchCORE,

• Online Collaborative Modelling
• Optimistic Concurrency Control
• Version Controlling
• Persisting the Models in Cloud data stores

Twitter Sentiment Analysiss

Data Collection:

• Twitter's Streaming API
• Keyword Filters: #StrangersThings, #Weather, #USAirlines

Tweets Pre-Processing:

• Conversion of Tweet texts into Lower-case
• Tokenizing the sentences (using NLTK Tokenizer)
• Removing Twitter Usernames & URLs (using RE)
• Using Stop Wordsavailable in English language dictionary

Tweets Labelling (Data Coding): CrowdFlower

Sentiment Analysis:

• SentiWordNet (Lexical Resource used for 'Opinion Mining')
• VADER - Valence Aware Dictionary and sEntiment Reasoner (Lexial and Rule-based Sentiment Analysis tool - Commonly used to analyze sentiments expressed in Social Media platforms)

Image Classification (Convolution Neural Networks)

• Classifying the images from Tiny Imagenet dataset down-sized to 64*64 and split across 40 categories
• Inception-v3, a CNN model that was pre-trained by Google on 100K images with 1000 categories was used
• 'Transfer Learning' technique was applied

Differential Privacy in Machine Learning

This project is a comparative study of Differential Privacy in Machine Learning. Two different Differential Privacy techniques to preserve the sensitive data are studied. They were applied on multiple data-sets like:

• MNIST (Modified National Institute of Stds. and Tech.)
• SVHN (Street View House Number)
• CIFAR-10 (Canadian Institute for Advanced Research)

Reddit Topic Classification

Supervised Machine Learning methods Random Forest & Stochastic Gradient Descent classifiers to classify short conversations extracted from Reddit. 8 Classes based on conversation topics (Hockey, Movies, NBA, News, NFL, Politics, Soccer and WorldNews).

Data Cleaning and Feature Extraction:

• Label Encoding (Fit and Transform) & Decoding (Inverse Transform) using Scikit-Learn Proceprocessing label Encoder
• Lemmatization (using WordNetLemmatizer) - NLTK Package (Done to increase the accuracy)
• Term Frequency-Inverse Document Frequency (TF-IDF) approach - feature weighting

Scraping Static and Dynamic Webpages

Two different types of web-pages (Static and Dynamic) were chosen and different strategies were used to scrape content from them. Also, a small analysis was done on the scraped data and the results are presented.

• Static Website: Who's Dated Who
• Dynamic Website: Empeopled and Boardest
• Who's Dated Who, despite being a static website, in order to fetch the URLs of all the web-pages, and the other two dynamic websites had Infinite Scrolling (where new posts are loaded dynamically when the user scrolls down to the bottom of the page)
• BeautifulSoup, Python library for screen scraping was made use of in static web-pages
• Selenium Web Driver was made use of to automatically control the browser and scroll until the last element loads

Twitter's Streaming & REST APIs

A comparative study on Twitter's REST and Streaming APIs and observing the Bias in Twitter Data Collection

• User's Popularity: "No. of Followers he/she possesses"
• Keyword Filter: '#Mexico', '#Earthquake'
• No. of Tweets collected using Streaming API: 50,375
• No. of Tweets from same set of users: 16,345
• No. of Unique User IDs: 34,025
• Sampling: 2 * 10000 User_IDs

Network Analysis

Analyzing Relationships in the Who-Dated-Who database, evaluate the abundance of overlapping dating partners and measuring the frequency of overlapping daters.

• Data Collection: Infinite Scrolling and Static Scraping
• Network Analysis: NetworkX & GEPHI
• NetworkX, a Python package was used for creation, manipulation and study of structure and dynamic of the network
• GEPHI, an Open-source Graph Visualization and Exploration tool was used

Component-based Distributed Travel Reservation System

A component-based distributed travel reservation system was developed using both TCP and RMI. More functionalities that were added on the RMI flavored system includes:

• Distribution and Scalability
• Transactions and Distributed Concurrency Control
• Reliability using 2-Phase Commit Protocol
• Performance Analysis

CrowdFlower - Sentiment Analysis

This project was carried out to analyze the sentiments expressed by users on Weather-related Tweets. It involves:

• Data Collection: Using Twitter's Streaming API
• Data Coding - Sentiment Labelling: Using CrowdFlower
• Data Analysis

Predicting Participant's Completion Times in 2017 Miami Marathon using Linear Regression

The dataset was collected from Athlinks and it consisted of detailed profiling of participants who have enrolled for Miami Marathon in the past, their finising times, etc.

• Feature Selection: Age (Continuous), Gender (Binary), AvgTime[Year] (Continuous) - Completion time in seconds for the entire marathon, AvgTimeForAllMarathons (Continuous) - completion time in seconds across all years with data for given participant, TotalNoOfRaces (continuous) - Equalivalent to Number of races participated prior to this Marathon
• Addressing Missing Data: (Done through Predictive Mean Matching) - By substituting the missing data with the person's mean (Average) time across all years

An efficient algorithm for Fast-Block Motion Estimation in High-Efficiency Video Coding

The algorithm involves a direction-based approach on several distinctly identified combinations of search points. The project was split into Prediction and Refinement modules • Five different schemes (EME 1 to 5) were introduced based on the combinations of search points and were experimented on a number of standard video sequences to predict the motion vector of the candidate block. The results justified that EME offered faster search minimizing search time

• Presented at the 4th IEEE International Conference on Recent Trends in Information Technology
• Published as a Book Chapter in "Emerging Technologies in Intelligent Applications for Image and Video Processing"

Miscellaneous Projects

Contains a list of websites and projects that have been carried out (during both Bachelor's and Master's)

• Dynamic and Responsive Websites: Designed eight websites using HTML5, CSS3 and JavaScript
• Modelling Re-usable concerns for singleton approaches
• Industry Project: NIMBUS Drive (File De-Duplication in Cloud Storage) - Unisys Cloud 20/20 V5
• Industry Project: Employee Record Management System (DIET, Puducherry, India)

Enhancement of latent fingerprints through Morphological Image Processing

Using Morphological Image Processing filters like 'Erosion', 'Dilation', 'Opening' and 'Closing' to enhance latent fingerprints.

• Presented and won Best Paper award in the National Conference on Bio-Inspired Systems and Technologies
• Published in the "International Journal of Engineering Research and Technology"

Dragon Connected Services (DCS)

During Summer Internship (2017): Worked in Dragon Connected Services (DCS): Cloud based service for providing content to the connected car. I was working on server-side development for weather-related content services.

Financial Services:

Client: Royal Bank of Scotland

Was a part of the Know Your Customer (KYC) and Client Due-Diligence team

Roles and Responsibilities:

• Client Interaction, requirements gathering and elicitation
• Creation of design prototypes
• Implementation using MVC Architectural pattern
• Responsible for problem, incident and change management for production bugs