Pradeep Bajracharya

PhD Student in Machine Learning

Rochester Institute of Technology

Biography

I am a final year PhD student in Computing and Information Sciences at Golisano College of Computing and Information Sciences, Rochester Institute of Technology. I am a Graduate Research Assistant under the supervision of Dr. Linwei Wang in Computational Biomedicine Lab.

Research Area

My primary research focus is on Deep Active Learning and its application in designing uncertainty-aware, efficient deep surrogate models for biomedical and scientific applications.

Technical Skills

Programming – Python, MATLAB, C++
ML Tools – PyTorch, Scikit-learn, TensorFlow
Tools & Libraries – MLFlow, WandB, Pandas, Numpy, Matplotlib, OpenCV, SLURM, Docker
Miscellaneous – AWS, AWS Sagemaker (Basics), LaTeX, Git

Interests

Deep Learning
Scientific Machine Learning
Active Learning
Uncertainty Quantification
Surrogate Modeling
Large Language Models
Natural Language Processing
Computer Vision

Education

PhD in Computing and Information Sciences

Rochester Institute of Technology
BE in Electronics and Communications Engineering, 2015

Pulchowk Campus, Institute of Engineering, Nepal

Publications

Meta-learning based Active Learning Approach for Computer-Assisted Pace-Mapping

Ventricular tachycardia (VT) is a major cause of sudden cardiac deaths, treated with catheter ablation that targets and destroys the tissue responsible for initiating or sustaining VT. Pace-mapping is a key practice for identifying ablation targets by matching stimulated sites’ ECG to the observed clinical VT. An AI-assisted pace-mapping approach can help clinicians identify the most likely target sites, reducing procedure time and increasing the success rate.

Pradeep Bajracharya, Dylan O’Hara, Casey Meisenzahl, Karli Gillette, Anton J Prassl, Gernot Plank, John L Sapp, Linwei Wang

On the interdependence between data selection and architecture optimization in deep active learning

Deep active learning (DAL) studies the optimal selection of labeled data for training deep neural networks (DNNs). While data selection in traditional active learning is mostly optimized for given features, in DNN these features are learned and change with the learning process as well as the choices of DNN architectures. How is the optimal selection of data affected by this change is not well understood in DAL.

Pradeep Bajracharya, Anton J. Prassl, Karli Gillette, Gernot Plank, Linwei Wang

On the interdependence between data selection and architecture optimization in deep active learning

Feasibility study on active learning of smart surrogates for scientific simulations

High-performance scientific simulations, important for comprehension of complex systems, encounter computational challenges especially when exploring extensive parameter spaces. There has been an increasing interest in developing deep neural networks (DNNs) as surrogate models capable of accelerating the simulations. However, existing approaches for training these DNN surrogates rely on extensive simulation data which are heuristically selected and generated with expensive computation – a challenge under-explored in the literature.

Pradeep Bajracharya, Javier Quetzalcóatl Toledo-Marín, Geoffrey Fox, Shantenu Jha, Linwei Wang

Active learning based Cardiac Tissue Parameter Estimation for Personalized model exploiting predictive uncertainty

Personalized cardiac models are crucial intervention tools for a multitude of cardiac health issues. As cardiac simulations become more complex and expensive, machine learning (ML) models demonstrated the potential to enable efficient model personalization and cardiac tissue parameter estimation.

Pradeep Bajracharya, Anton J. Prassl, Karli Gillette, Gernot Plank, Linwei Wang

Active learning based Cardiac Tissue Parameter Estimation for Personalized model exploiting predictive uncertainty

Fast Posterior Estimation of Cardiac Electrophysiological Model Parameters via Bayesian Active Learning

We present a Bayesian active learning method to directly approximate the posterior pdf function of cardiac model parameters, in which we intelligently select training points to query the simulation model in order to learn the posterior pdf using a small number of samples. We integrate a generative model into Bayesian active learning to allow approximating posterior pdf of high-dimensional model parameters at the resolution of the cardiac mesh.

Md Shakil Zaman, Jwala Dhamala, Pradeep Bajracharya, John L. Sapp, B Milan Horácek, Katherine C Wu, Natalia A Trayanova, Linwei Wang

Fast Posterior Estimation of Cardiac Electrophysiological Model Parameters via Bayesian Active Learning

Semi-supervised Medical Image Classification with Global Latent Mixing

In this work, we argue that regularizing the global smoothness of neural functions by filling the void in between data points can further improve SSL. We present a novel SSL approach that trains the neural network on linear mixing of labeled and unlabeled data, at both the input and latent space in order to regularize different portions of the network.

Prashnna K. Gyawali, Sandesh Ghimire, Pradeep Bajracharya, Zhiyuan Li, Linwei Wang

Embedding High-dimensional Bayesian Optimization via Generative Modeling - Parameter Personalization of Cardiac Electrophysiological Models

We present a novel concept that uses a generative variational auto-encoder (VAE) to embed HD Bayesian optimization into a low-dimensional (LD) latent space that represents the generative code of HD parameters. We further utilize VAE-encoded knowledge about the generative code to guide the exploration of the search space.

Jwala Dhamala, Pradeep Bajracharya, Hermenegild J Arevalo, B Milan Horácek, Katherine C Wu, Natalia A Trayanova, Linwei Wang

Indoor Odometry and Point Cloud Mapping

Indoor localization and mapping is an important problem with many applications such as emergency response, architectural modeling, and historical preservation. In this project, a metrically accurate, GPS-denied, indoor 3D static mapping system was developed using a moveable base coupled with three degree of freedom IMU.

Prabhat Sanu Ligal, Bikram Acharya, Pradeep Bajracharya, Prasun Shrestha, Pratik Pokharel, Sharad Kumar Ghimire

Professional Experience

Graduate Research Assistant

Computational Biomedicine Laboratory, RIT

Aug 2018 – Present Rochester, New York

Developed meta-learning based active neural surrogate model for localization of source of Ventricular Tachycardia on heart, achieving over 70% improvement over Bayesian Optimization.
Developed a locally accurate neural surrogate for inverse cardiac tissue parameters estimation responsible for irregular ECG signals, improving localization error by 71% over passive baselines.
Developed active learning-based surrogate model for scientific simulations in an application of oxygen diffusion in retina as an cost-effective alternative.
Explored interdependence of active learning and deep learning architectures (RESNETs to LLMs like BERT, DistilBERT and RoBERTa). Architecture optimization (via Neural Architecture Search) found to be more important than data optimization.
Developed novel gradient-based acquisition function for faster and efficient phase mapping for scientific simulations in an application of a colloidal system, reducing simulation time by 92%

Graduate Teaching Assistant

Chester F. Carlson Center for Imaging Science, RIT

Aug 2018 – May 2019 Rochester, New York

COS-IMGS-111 - Image Science Fundamentals
COS-IMGS-180 - Intro to Computing and Control

Senior Developer

Kazi Studios

Aug 2016 – Jun 2018 Bhanimandal, Lalitpur, Nepal

Development of Web based solutions, and CRM systems including but not limited to medical inventory system, tourism portals and android applications. Also worked on smart home system controlled via android, and IOS platform and chatbot integrated in messenger.

Teaching Assistant

Pulchowk Campus, Tribhuvan University

Apr 2016 – Aug 2016 Pulchowk, Lalitpur, Nepal

In charge of conducting “Basic Electronics Lab” sessions for first year students.

System Engineer

E&T Nepal Pvt. Ltd.

Nov 2015 – Apr 2016 Lokanthali, Bhaktapur, Nepal

Development of Calculation Solver for CFD simulation for automobile application
Technology used: CUDA Programming, C++.

Intern

E&T Nepal Pvt. Ltd.

May 2014 – Dec 2014 Lokanthali, Bhaktapur, Nepal

Led research project named High Speed Data Transfer to improve speed of existing data transfer.

Awards and Scholarships

MICCAI 2021 Satellite Event Participation Grant

MICCAI2021 Sep 2021

MICCAI Society provided 50 Satellite Events Participation Grants which included a member registration fee for the MICCAI Satellite Events, in conjunction with a one-year MICCAI Society membership fee.

Prof. F.N. Trofimenkoff Academic Achievement Award

Tribhuvan-University Nov 2019

Awarded for graduating top of the class (2015) in BE Electronics and Communications Engineering, Pulchowk Campus.

RIT Ph.D. Merit Scholarship

RIT Aug 2018

Financial assistance for Ph.D. studies at Rochester Institute of Technology since August 2018