Avatar

Nirmal Praveen Suthar

Sophomore Undergraduate

“Dream it. Wish it. Do it.”

Indian Institute of Technology Kanpur

About Me

Hey, I’m a Sophomore at Indian Institute of Technology, Kanpur majoring in Computer Science and Engineering, broadly interested in the areas of Deep learning and applied machine learning. I am genuinely interested in contributing to open source, Machine learning ecosystem. .

My sophomore year research work is centred around exploring new and challenging dataset MemexQA. Under the supervision of Badri Narayan Patro at delTA Lab IIT Kanpur, We developed a VQA based model to tackle real-life question answering problems on multimedia collections such as personal photo albums.

Lately, I've been fascinated by the application of Generative Adversarial Network to counter digital manipulation techniques. Being able to model the real world and generate realistic data has many exciting applications. When such data is generated with an intent to deceive the user (or the computer system) into believing that the fake data being generated is real, and these attacks span across multiple media including image, speech & audio, and text.

Last summer, I spent some time exploring Generative Adversarial Network and multi-modal GANs, in particular. I was also involved in SOC, IITK, where are team, developed a web-based platform for managing documents and application and proceedings made by the concerned users. I’ve had some exposure to NLP and its applications , though I’m much less acquainted with the subject that I would like to be.

Interests

  • Computer Vision
  • Natural Language Processing
  • Generative Adversarial Network
  • Probabilistic Machine Learning
  • Representation Learning
  • Object Detection

Education

  • BTech in Computer Science and Engineering, 2018-Present

    Indian Institute of Technology Kanpur, India

Projects

*

MemexQA Dataset

developed a VQA based model to tackle real-life question answering problems on multimedia collections such as personal photo albums

A study in GANs

Studied Convolutional nueral networks and Generative adversarial networks in-depth . Implemented and played with DCGAN, class conditioned GANs like ACGAN,CGAN,InfoGAN and Style Transfer GANs like DiscoGAN, CycleGAN and StarGAN . Implemented dataloader and progress bar feature in TorchGAN(research framework to train GANs) and tried implementing a YAML parser to train GANs automatically .

VoteChain - Voting app using Blockchain

Made a voting app (Microsoft codefundo++ Hackathon 2019) using Microsoft Azure Blockchain services for a more secure election.

OfficeMarshall - SOC,IITK

Developed a web-based platform for managing documents and application and proceedings made by the concerned users.

Attention Based Models for Visual Question Answering

Studied the theory of encoder decoder sequence models, soft attention and hard attention and reproduced the results in Show, Ask, Attend, and Answer

Explainable Machine Learning

Studied LIME, technique that explains the predictions of any classifier in an interpretable and faithful manner. "Why Should I Trust You?": Explaining the Predictions of Any Classifier

Contact