2025
FLAME: Federated Learning with Masked Autoencoders and Mean-Prototypes Embedding for Sparsely Labeled Medical Images
M. K. Kundalwal and D. Mishra · IEEE Trans. Emerging Topics in Computational Intelligence
About
Biography
I am a PhD Candidate in the Department of Computer Science & Engineering at the Indian Institute of Technology Jodhpur, under the supervision of Dr. Deepak Mishra. My research spans federated learning, trustworthy AI, privacy-preserving machine learning, and medical computer vision.
My doctoral research focuses on developing robust and privacy-centric distributed learning pipelines tailored for clinical settings. I have proposed four novel FL frameworks:
FLAME
Self-supervised FL with masked autoencoders for sparse labels (5% data)
AR2FL
Byzantine-resistant aggregation via mean latent representations
BAFL
Communication-efficient FL through base-layer parameter ablation
SPU
Selective parameter unlearning for GDPR-compliant model updates
Education
Academic Path
PhD
2019–25
Computer Science & Engineering
Indian Institute of Technology Jodhpur
M.Tech
2016–18
Computer Science & Engineering
NIT Patna · CGPA 8.24/10
B.Tech
2012–16
Computer Science & Engineering
Poornima University Jaipur · CGPA 7.38/10
Research
Focus Areas
01
Federated Learning
Lightweight, communication-efficient distributed training with client contribution normalization and base-layer parameter ablation across heterogeneous channels.
02
Federated Unlearning
Algorithmic frameworks for data unlearning in distributed systems, complying with GDPR's "Right to be Forgotten" without full retraining.
03
Trustworthy AI
Mitigating Byzantine and adversarial poisoning attacks through robust servers that evaluate client similarity using latent representation alignment.
04
Medical Image Analysis
Automated clinical diagnostics under sparse annotation using self-supervised architectures for classification and segmentation.
Publications
Selected Work
Journal Articles
2025
AR2FL: Anomaly-Resistant Robust Framework for Federated Learning
M. K. Kundalwal and D. Mishra · IEEE Trans. Artificial Intelligence
2019
An Improved Privacy Preservation Technique in Health-Cloud
M. K. Kundalwal, K. Chatterjee, A. Singh · ICT Express, Elsevier
Conference Proceedings
2026
Federated Model Synchronization for Diagnostic Redefinition through Selective Parameter Unlearning
M. K. Kundalwal and D. Mishra · WACV 2026 · 🏆 DEI Travel Grant
2024
Client Contribution Normalization for Enhanced Federated Learning
M. K. Kundalwal, A. Saraswat, I. Mishra, D. Mishra · IEEE INDICON
2022
BAFL: Federated Learning with Base Ablation for Cost Effective Communication
M. K. Kundalwal, A. Saraswat, I. Mishra, D. Mishra · ICPR 2022
2018
A Privacy Framework in Cloud Computing for Healthcare Data
M. K. Kundalwal, A. Singh, K. Chatterjee · ICACCCN
Experience
Research Roles
Senior PhD Researcher
IIT Jodhpur · 2019 – 2025
Led research in distributed computer vision frameworks. Developed PyTorch pipelines for federated unlearning, robustness estimation, and self-supervised autoencoding. Published in IEEE Transactions.
Project Associate — Radar Systems
IIT Jodhpur · Sponsored Project
Developed ML algorithms for radar signal classification, clutter removal, and moving-target tracking for defense research.
Skills
Technical Stack
Core Methods
Federated LearningMachine UnlearningByzantine RobustnessSelf-Supervised LearningTrustworthy AIComputer Vision
Frameworks & Tools
PyTorchTensorFlowCUDANumPyScikit-LearnGitDockerLaTeX
Languages
PythonC++SQLMATLAB