Shasvat-Desai

ABOUT ME

Graduate student of Computer Science at University of Massachusetts Amherst, My areas of interests include the application of Machine Learning and Computer Vision. Currently, I have been working on the task of Pedestrian Detection in Multispectral Images and recently, my paper got accepted in a workshop at CVPR 2019. My comprehensive resume is available here.

NEWS

April 2019 - Will start working as a Software Engineer(Computer Vision) at Orbital Insights,Boston MA,

April 2019 - Our paper Pedestrian Detection in Thermal Images using Saliency Maps accepted to the IEEE Workshop on Perception Beyond the Visible Spectrum at CVPR, 2019 (Long Beach, California). Authors:Debasmita Ghose,Deep Chakrobarty,Sneha Bhattachrya,

Summer 2018 - Received the DAAD RISE Professional Scholarship - 2018 to pursue a research internship with BayerBusiness Services GmBH, Colonge, Germany,

Fall 2017 - Started Masters in Computer Science at University of Massachsetts, Amherst,

EXPERIENCE

Research Assistant at the Information-Fusion-Lab

College of Information and Computer Sciences, UMass Amherst                       Sept 2018-Present

Worked on the task of Pedestrian detection in Multispectral Images.

I was advised by Dr.Madalina Fiterau and Dr. Tauhidur Rahman

Recently, our paper got accepted in PBVS workshop at CVPR 2019 for the novel work on the use of Saliency Maps, generated using Deep Saliency network, along with the thermal images to improve the daytime performance on thermal images for the task of Pedestrian Detection.

Natural Language Processing Research Intern

Bayer Business Services, Leverkusen, Germany                                                  June 2018-August 2018

• Identifying use cases of pharmaceutical products of Bayer using classification algorithms on their text corpus of medical data
• Prototyping the development of a Generative Chatbot model to automate the process of answering the questions asked by medicine practitioners, pharmacists, pharamaceutical sales representatives and patients.

Deep Learning Research Assistant

School of Computer and Information Sciences, UMass Amherst, USA                      Spring 2018
Advisor: Prof. Eliot B. Moss

• Analysed waveforms of memory traces generated from cache memory accesses of the program to model the underlying behavior of computer programs under Professor Eliot Moss
• Applied Sequence Modelling techniques- RNN, LSTM, Sequential autoencoders and Statistical models- ARIMA to model the time dependent nature of the memory traces and cluster similar computer programs

Computer Vision Intern

Wipro Technologies, Pune, India                                                                           May 2016– July 2016
Used C# ,C++ and OpenCV to develop modules for the Image Processing software used in the LEXT Industrial Confocal Laser Microscope

SKILLS

• Languages:

  • Java

    Python

    C

    C++

    MATLAB

    C#

    R

    Javascript

    HTML

    CSS

• Frameworks:

  • Keras

    PyTorch

    TensorFlow

    .NET

    Hadoop

    MapReduce

    Git

EDUCATION

University of Massachusetts, Amherst                                                                         Fall 2017-Fall 2019
Masters of Science, Computer Science                                                                              GPA:3.95/4.0
Relevant Courses: Deep Learning, Computer Vision, Machine Learning, Algorithms of Data Science, Intelligent Visual Computing (3D Computer Vision), Natural Language Processing, Reinforcement Learning

Veermata Jijabai Technological Institute (V.J.T.I), Mumbai, India                         Fall 2013-Fall 2017
Bachelor of Technology, Computer Science                                                                     CGPA:8.65/10
Relevant Courses: Network Management Systems, Engineering Mathematics, Discrete Mathematics, Artificial Intelligence and Data Mining, Algorithms, Cloud Computing, Operating Systems, Web Development

PROJECTS

Telekinesis - A Multi User and Multi Class Classification of EEG data

Code | Report                                                                       May 2018-Present

• Using a cascaded and parallel Convolutional Recurrent architecture using the learned representation from Variational Autoencoders to generalizing the EEG data for various multiuser and multiclass classification instead of training a model on a subject specific data

Analysis of Computer Program Behaviors using Sequence Modelling Techniques

Report                                                                       Spring 2018

• Performed a comparative analysis of program performance on different complilers
• Predicted the next blocks of cache memory accessed by the program based on the nature of memory access in the past
• Used LSTM and sequential autoencoders to model the time sequence data generated from cache memory traces.

A Comparative Study of Architectures for 2D Image Segmentation

Code | Report                                                                       Spring 2018

• Performed a comparative study of various Deep Learning models- FCN, U-Net, Dilated Convolutions, Dense Nets by modifying them for Image Segmentation in Keras Framework on PASCAL VOC dataset
• Utilized the power of transfer learning techniques and achieved improvement in the performance in U-Net

Question Answering on SQuAD

Code | Report                                                                       Spring 2018

• Developed a novel architecture through modification of the BiDAF Network using PyTorch framework on the Stanford Question Answering Dataset and Spacy library
• Leveraged the power of self attention used in Tranformer Networks and exploited the Dependency Parse structure of the text to enhance the performance of the model
• Achieved an F1 score of 72.14 by adding a dependency parse layer, implemented with transformer, to BiDAF - an improvement over AllenAI’s BiDAF model’s score of 71.49

Automatic Generation of Highlights from a Sports Video

Code | Report                                                                       Fall 2017

• Automated the task for generating highlights of a game by recognizing audience reactions to the events during the game play, using a 3D Convolutional Neural Networks in the Keras framework on the SHOCK Dataset
• Experimented with transfer learning techniques by Fine tuning the models pretrained on action recognition datasets

ACHIEVEMENTS

• Awarded the DAAD RISE Professional Scholarship 2018- Opportunity to intern at the headquarters of Bayer Business Services Research laboratory in Germany as an NLP intern for Summer 2018

CONTACT