Karan Jeswani

I'm a

About

I'm a Graduate Student in the Indian Institute of Science (IISc), Bangalore. I am in the Medical Imaging Group, in the Department of Computational and Data Science (CDS). I am currently working on Medical Image Synthesis using conditional Generative Adversarial Network, by posing it as an Image to Image translation problem in Computer Vision.

Interests

  • I’m interested in all things related to Mathematics, and Programming, so Computational Sciences comes really at the intersection of these.
  • I am interested in understanding, applying and improving deep neural network architectures, for computer vision tasks.
  • I do like to create several patterns and nice looking animations for educational purposes or art! I do this using these fantastic creative coding libraries called p5.js and processing. This is how I was introduced to creative coding and I loved it as soon as I saw it.

Skills

Programming Languages


Python PySpark C C++ MPI OpenMP Using GPU accelerators with CUDA JavaScript

Deep Learning Libraries


PyTorch TensorFlow Keras

Numerical Computing


Numpy Scipy Pandas MATLAB

Visualization Libraries


Matplotlib SimpleITK ParaView Processing p5.js FSL and ITK-SNAP

Resume

Sumary

Karan Jeswani

I’m a Graduate Student, in the Department of Computational and Data Sciences at the Indian Institute of Science, Bangalore working with Professor Phaneendra Yalavarthy in the Medical Imaging Group

Education

M. Tech Computational Data Science

2019 - Present

Indian Institute of Science (IISc), Bangalore, India

B. Tech Mechanical Engineering

06/2013 - 05/2017

Veermata Jijabai Technological Institute (VJTI), Mumbai, India

  • CGPA: 8.40 / 10, Ranked 17th in a class of 70
  • Lab: Statistical Thermodynamics

Professional Experience

Production Engineer, Power Transformers Division

06/2017 - 12/2017

CG Power & Industrial Solution Ltd., Mumbai, India

  • Succesfully implemented IEC 60076 standards as quality baseline in manufacturing and testing of Power Transformers.
  • Supervised the Core-Coil Assembly and Tanking operations for Power Transformers.

Projects

MRI to CT Image Modality Translation using Conditional GAN

01/2020 - 06/2020

Department of Computational and Data Sciences, IISc, Bangalore, India

  • Implemented cGAN, commonly used for Image-to-Image Translation in Computer Vision (CV), to synthesize CT scan images given MRI-T1 images. Used smaller network that still achieves performance comparable with State-of-the-Art.
  • Performed hyper-parameter tuning for both networks using grid search strategy for extracting best performance, in terms of PSNR and SSIM, for custom dataset.
  • Observed that adding perceptual loss from VGG-19 and incorporating neighbouring slices improves network performance.

Automatic Evaluation Metrics for Multi-Sentence Text

01/2020 - 06/2020

Department of Computer Science and Automation, IISc, Bangalore, India

Conducted as a part of Machine Learning Course

  • Implemented several common metrics in Natural Language Processing (NLP) like BLEU, ROUGE, BERTScore, Sentence Movers Similarity for evaluating single and multi-sentence text on 3 datasets, namely WMT-18, CNN News, Student Answer Scripts.
  • Performed experiments with various embeddings like GloVe, ELMo, BERT, RoBERTa, with all types of metrics.
  • Cosine Similarity based metrics provided better correlation with humans as compared to metrics based on word matching or Wasserstein distance, for both single-sentence & multi-sentence text.

Particle Filter for Video based Object Tracking

01/2020 - 06/2020

Department of Computational and Data Sciences, IISc, Bangalore, India

Conducted as a part of Parallel Programming Course

  • Applied the popular state estimation technique called Particle Filter for video based object tracking, and parallelized the algorithm to work on the GPU using OpenCV and C++.
  • Improved the Parallel Resampling algorithm by exploiting non-decreasing property of cumulative distribution function.
  • This increases latency for video based object tracking, since it uses faster technique for resampling

Design and Analysis of Physical Vapour Deposition system using DSMC for a Binary Mixture of Gases

05/2016 - 05/2017

Mechanical Engineering Department, VJTI, Mumbai, India

Final Year B. Tech Project

  • Performed DSMC simulation of binary mixture of gases, tracking its motion and collisions at low pressure in 3D, using C++.
  • Empirically calculated parameters to maintain flow rate of the primary component in the mixture.

Portfolio

Being in CDS, in IISc, I got an amazing opportunity to do lots of courses and projects along with them. Here is a brief idea about what I did and the conclusion (for you to learn all the things in a condensed manner).