Rohit Kumar Jena

I am a first year Masters student at The Robotics Institute, Carnegie Mellon University advised by Prof. Katia Sycara.

I completed my bachelors in Computer Science and Engineering from Indian Institute of Technology, Bombay in 2019. My undergraduate thesis is based on Perfect Sampling and Uncertainty Estimation in Deep Networks where I was advised by Prof. Suyash P. Awate.

Email  /  CV  /  Google Scholar  /  ArXiv /  LinkedIn


  • [Aug 2019]   Graduated from IIT Bombay with a Bachelors in Computer Science and Engineering with Honors.
  • [Aug 2019]   Received the Research Excellence Award for outstanding research work done during undergraduate studies.
  • [May 2019]   Awarded Travel Grant from the C'1992 Legacy Project Funds to present at IPMI 2019, Hong Kong.
  • [May 2019]   paper on perfect sampling for Bayesian MRFs accepted at Medical Image Analysis journal.
  • [Feb 2019]   paper on uncertainty estimation selected for oral presentation and opening talk of IPMI 2019.
  • [Feb 2019]   paper on uncertainty estimation and calibration for neural networks accepted at IPMI 2019, Hong Kong.


I'm interested in computer vision, machine learning, and medical image analysis. Some of my work has focused on uncertainty estimation in the context of perfect sampling for deep MRFs, and fast aleatoric uncertainty estimation and calibration for neural networks.

segBayes A Bayesian Neural Net to Segment Images with Uncertainty Estimates and Good Calibration
Rohit Jena, Suyash P. Awate
International conference on Information Processing in Medical Imaging (IPMI) 2019
Oral presentation, opening talk of the conference, acceptance rate ~10%
paper   /   slides

We propose a novel Bayesian decision theoretic deep-neural-network (DNN) framework for image segmentation, enabling us to define a principled measure of uncertainty associated with label probabilities. Moreover, our framework leads to a novel Bayesian interpretation of the softmax layer. We propose a novel method to improve DNN calibration.

perfectmcmc Estimating uncertainty in MRF-based image segmentation: A perfect-MCMC approach
Suyash P. Awate, Saurabh Garg, Rohit Jena
Medical Image Analysis (MedIA) 2019, 55:181-196, Elsevier

We propose the modern paradigm of perfect MCMC sampling to sample multi-label segmentations from generic Bayesian Markov random field (MRF) models, in finite time for exact inference. Furthermore, for exact sampling in generic Bayesian MRFs, we extend the theory underlying Fill's algorithm to generic MRF models by proposing a novel bounding-chain algorithm.

Selected Projects
19 Variational Autoencoder with Arbitrary Conditioning (unofficial implementation)

A PyTorch implementation (unofficial) of the ICLR 2019 paper Variational Autoencoder with Arbitrary Conditioning. Developed as part of the ICLR reproducibility challenge.

obj Objects that Sound (unofficial implementation)
code /  report

A re-implementation of Deepmind's paper Objects that Sound using PyTorch. Developed as part of the Advanced Machine Learning graduate course at IIT Bombay.

cfd Crop Disease Detection

Leveraged transfer learning from ImageNet pre-trained models to identify from among 38 classes of plant-disease pairs. Also trained an auxiliary network which uses name embedding of the crop and the image to produce more accurate results.
Secured Second Runners-up position in Code.Fun.Do hackathon by Microsoft.

iitb Undergraduate Teaching Assistant, CS101 Spring 2018

Undergraduate Teaching Assistant, CS215 Fall 2018

Undergraduate Teaching Assistant, CS251 Fall 2017

Undergraduate Teaching Assistant, MA105 Fall 2016

Thanks to Jon Barron. for sharing this awesome template!!