Anurag Roy

Anurag Roy

pursuing PhD in Multimodal Machine Learning

CSE, IIT Kharagpur


I graduated from the Computer Science and Technology department(with Hons.) of IIEST Shibpur in the year 2017. I worked as a software engineer at Polaris Networks for a year(July 2017 - Jul 2018). Currently I am a Ph.D. scholar at IIT Kharagpur under the supervision of Prof. Saptarshi Ghosh of Computer Science and Engineering Department. I am also a member of the Complex Networks Research Group(CNERG)


  • Machine Learning
  • Multimodal Learning
  • Information Retrieval


  • PhD in Computer Science and Engineering, 2019

    Indian Institute of Technology Kharagpur

  • B.E.(Hons) in Computer Science and Technology, 2017

    Indian Institute of Engineering Science and Technology, Howrah, West Bengal

  • CBSE XII Science with Computer Application, 2013

    Army Public School, Ballygunge, Kolkata, West Bengal


Programming Skills









Senior Research Fellow

CNeRG Lab, CSE Department, IIT Kharagpur

Jul 2020 – Present Kharagpur, West Bengal

Applied Scientist Intern

India ML Team, Amazon Development Center India Pvt. Ltd.

May 2019 – Jul 2019 Bengaluru, Karnataka
– Predicted Product Trust Score – Developed machine learning models to Predict Trust Score

Ph.D. Research Scholar

CNeRG Lab, CSE Department, IIT Kharagpur

Jan 2019 – Present Kharagpur, West Bengal
Under the Supervision of Prof. Saptarshi Ghosh

Junior Research Fellow

CNeRG Lab, CSE Department, IIT Kharagpur

Jul 2018 – Jul 2020 Kharagpur, West Bengal

Software Developer

Polaris Networks

Jul 2017 – Jul 2018 San Jose, California

Responsibilities include:

  • Development of 5g core network stack
  • Development of MCPTT tester

Research Intern

IIT Kanpur

May 2017 – Aug 2017 Kanpur, Lucknow


Responsibilities include:

  • Understaning the different kind of morphological variations existing over different languages.
  • Developing of an un-supervised clustering algorithm to capture the morphological variants of a word.

Research Intern

IIEST Shibpur

Dec 2016 – Jan 2017 Howrah, West Bengal
Developed using python and scikit-learn a program which identified rumor tweets in real-time by learning from previous data.

Research Intern

CNeRG Lab, CSE Department, IIT Kharagpur

May 2016 – Jul 2016 Kharagpur, West Bengal


Responsibilities include:

  • Developin an on-line survey application in flask and jinja.
  • Using machine learning models to evaluate accuracy

Undergraduate Student

IIEST Shibpur

Aug 2013 – May 2017 Howrah, West Bengal
BE in Computer Science and Technology

Recent Publications

Quickly discover relevant content by filtering publications.

ZSCRGAN: A GAN-based Expectation Maximization Model for Zero-Shot Retrieval of Images from Textual Descriptions

Most existing algorithms for cross-modal Information Retrieval are based on a supervised train-test setup, where a model learns to …

Distributed Representation of Tags for Active Zero Shot Learning [short paper]

Extreme multi-labeled classification (XMLC) refers to the problem of tagging items to its most relevant subset of class labels from an …

An Unsupervised Normalization Algorithm for Noisy Text: A Case Study for Information Retrieval and Stance Detection

A large fraction of textual data available today contains various types of ‘noise’, such as OCR noise in digitized …

Retrieving Information from Multiple Sources [poster]

The Web has several information sources on which an ongoing event is discussed. To get a complete picture of the event, it is important …

Combining Local and Global Word Embeddings for Microblog Stemming [short paper]

Stemming is a vital step employed to improve retrieval performance through efficient unification of morphological variants of a word. …

Recent Posts


Summary While Existing methods on few-shot image segmentation focus on 1-way segmentation, this paper focuses on k-way segmentation tasks. Existing Few-shot learning algorithms suffer from: Distribution Divergence: Most existing methods require to be pre-trained on ImageNet.

From CPP to Java

Some differences between c++ and java: Java compiled code is platform independent whereas c++ compiled code is platform dependent Java interpreter reports the run-time error that caused the execution to halt unlike in c/c++ programs which may simply crash

Stats 101

Sampling Theory Data scientists are required to draw conclusions about a group, a.k.a population from a few samples of it because getting the entire population is intractable. This process of drawing samples is called sampling.

Sampling Theory and Distributions

Sampling Theory Data scientists are required to draw conclusions about a group, a.k.a population from a few samples of it because getting the entire population is intractable. This process of drawing samples is called sampling.

How to Install Packages Locally in Linux

Some Prerequisites What is a variable? A variable is a storage location for a value. Linux has environment variables. It can store strings, numbers , etc. just like the variables in C, C++, python, or any other programming language.


  • 7278389228
  • Dept. of Computer Science and Engineering, IIT Kharagpur, Kharagpur, West Bengal 721302