Soumyajit Ganguly

Graduate Researcher,
budding Data Scientist

© 2017. All rights reserved.

Soumyajit Masters (MS by research) in Computer Science at IIIT-Hyderabad, India. I am affiliated to Data Science and Analytics Centre (DSAC) and advised by Dr. Vikram Pudi.

My present research work is about exploring the use of full-text information for a richer representation and understanding of academic literature and citation networks in Computer Science domain. Apart from this, I am interested in solving various real world problems with machine learning techniques.

Publications
  • Paper2vec: Combining Graph and Text Information for Scientific Paper Representation. European Conference on Information Retrieval, ECIR 2017. PDF

  • Author2Vec: Learning author representations by combining content and link information. World Wide Web Conference, 2016. PDF

  • Competing algorithm detection in scholarly articles. iKDD Conference on Data Science, 2016. PDF

  • Stroboscopic image in smartphone camera using real time video analysis. SPIE Electronic Imaging, Enabling Technologies in Multimedia, 2015. PDF

  • A Novel Approach to Design a RealTime Control of HandGesture Mimicking Robotic Movements on an iRobot Create. IOSR Journal of Computer Engineering (IOSRJCE) vol 16 issue 3, 2014.

Work Experience
Tech BLA Solutions. Kolkata, India
Jun 2012 - May 2014

 Computer Vision Engineer
I worked as a computer vision engineer for the iOS platform. My job was implementing back-end algorithms and their required updates and maintenance. Some specifics about my work over there include: Hand Gesture Recognition, Motion Analysis from video, Digital Image manipulation and enhancements.

Education
Master of Science
(MS by research)
2014 - present		 International Institute of Information Technology (IIIT) - Hyderabad
Computer Science
Bachelor of Technology (B.Tech.)
2008 - 2012		 West Bengal University of Technology - Kolkata
Electronics and Communication

Projects

Aug 2016 - Oct 2016
Python, Keras
Deep Learning
 DeepSecure: User Identification via Keystroke Patterns
Implemented deep neural network architectures to learn interactions between keystroke patterns thus helping in user identification and user recognition.

Dec 2015 - Aug 2016
Python, Torch
Deep Learning
 Academic author embedding using Link and Text information and Representaion Learning
Trained a neural network to get vector embeddings for authors using negative sampling similar to in word2vec to get novel feature representations for both authors and papers in academic networks.

May 2015 - Dec 2015
Python, NLTK
Text Mining
 Mining Competing Algorithms from Scholarly Articles
Used unsupervised techniques to extract out competing algorithm names and proposed a novel approach for algorithm name and paper reference number linking in Computer Science papers.

Mar 2015 - May 2015
Python, Scikit-learn
Web mining
 Web Artifact scoring in Social Networks
Implemented search engine on Wikipedia from scratch. Extracted text content from social artifact links using python crawlers. For non-linked artifacts, applied k-means clustering and k-nearest neighbor matching to get top articles from Wikipedia. Computed scores based on likes, shares, retweets, favorites etc per artifact.

Sep 2014 - Dec 2014
Python, Torch, Scikit-learn
Deep Learning
 Object Recognition, Image Classification
Performed Image classification on the CIFAR-10 dataset. Used SIFT, HOG features with visual bag of words techniques and a SVM classifier to create a baseline. Then used Convolutional Neural Networks to achieve state-of-the art in classification performance.

Aug 2014 - Nov 2014
Python, Scikit-learn
Data Mining
 Bike Sharing Demand
Used supervised machine learning techniques like Random Forests and Gradient Boosted Trees to predict the number of bike riders on a given day based on historical climate and logistic data.

Mar 2013 - Apr 2014
Objective-C, Open-CV
Computer Vision
 Insta Fusion, Wrap Cam, Wrap Frame
Implemented touch based algorithms for iOS devices to perform basic image manipulations, enhancements. Implemented a touch based Image warping feature and digital stroboscopy which creates a motion trail photo for moving objects in given video clip.

Jun 2012 - Mar 2013
Objective-C, Open-CV
Computer Vision
 Box the Bag
Used supervised learning techniques to detect hand skin color for video segmentation and then detect circular hand motion from the segmented video, all in real-time front camera enabled smartphone devices.

Hobbies
  • Listen to instrumentals, try to play them on the Keyboard
  • Learn new subjects through online courses, youtube videos
  • GPU enthusiast, Play video games
    •

Contact

Email: soumyajit.ganguly@research.iiit.ac.in, soumyajit.ganguly@ymail.com

CDE Lab,
IIIT-Hyderabad,
Gachibowli, Hyderabad
India - 500032