Parth Sharma
Parth Sharma
PhD Research Scholar
PhD Research Scholar
Embedded Files
Indian Institute of Technology, Roorkee
A Ph.D. researcher in distributed and federated learning for communication systems, focusing on divergence-based adaptive algorithms, communication-efficient training, and secure decentralized frameworks.
Programming
Programming
Python, MATLAB, C, LaTeX
Skills
Skills
Research Interest
Research Interest
AI for 6G, Distributed & Decentralized Learning, Communication-Efficient Federated Learning, Secure & Privacy-Preserving Learning (QKD/FL), Channel Modeling & Estimation, IoT & Edge Intelligence, Graph & Multimodal Learning
Software Packages
Software Packages
PyTorch, Hugging Face (transformers, torchvision, timm), Sionna, SionnaRT, OpenCV, scikit-learn, pandas, NumPy, Matplotlib, seaborn
Recent Publication
Recent Publication
Parth Sharma, and Pyari Mohan Pradhan
12 November 2025 • 2025 IEEE 2nd International Conference on Communication Engineering and Emerging Technologies (ICoCET)
This paper introduces a secure and decentralized framework for facial keypoint detection in multi-agent systems by integrating a diffusion-based learning strategy with Quantum Key Distribution (QKD). The proposed QKD-enhanced diffusion strategy ensures robust data privacy and protection against model inversion and unauthorized data reconstruction attacks, achieving strong accuracy and secure communication during collaborative learning. Extensive simulations on the Landmark guided face Parsing (LaPa) dataset validate the effectiveness of the approach across various real-world scenarios. Compared to other strategies, the proposed QKD-enhanced diffusion strategy achieves an optimal balance between performance and data privacy. The results highlight the adaptability of the proposed approach, establishing a robust foundation for secure, distributed learning, designed to operate in environments potentially exposed to classical and quantum-based security threats.
Parth Sharma*, Prem Chand Panwar*, and Pyari Mohan Pradhan (* Equal Contribution)
12 November 2025 • 2025 IEEE 2nd International Conference on Communication Engineering and Emerging Technologies (ICoCET)
The integration of advanced adaptive filtering algorithms with real-time hardware platforms has transformed the potential of wireless sensor networks (WSNs) for environmental monitoring and safety-critical applications. This paper presents a comprehensive study on the implementation and performance evaluation of distributed least mean squares (LMS) algorithm variants, including LMS, Beta divergence based LMS (BLMS), and Kullback-Leibler divergence based LMS (KL-LMS), using hardware platforms. The study explores various distributed strategies, including incremental, consensus, combine-then-adapt (CTA) diffusion, and adapt-then-combine (ATC) diffusion techniques, to enhance estimation accuracy and network scalability. A NodeMCU ESP8266 microcontroller based setup, integrated with DHT11 and BME280 sensors for temperature and environmental monitoring, demonstrates the practical applicability of these strategies. Real-time data transmission between nodes is facilitated using the lightweight MQTT protocol. Hardware experiments reveal that BLMS, when paired with the consensus strategy, achieves superior performance in terms of mean square error (MSE), making it ideal for dynamic and resource constrained environments. These findings underscore the efficacy of adaptive filtering techniques in improving WSN reliability and scalability, providing a robust foundation for applications in smart cities, environmental monitoring, and safety systems.
Shekhar Pratap Singh, Parth Sharma, and Pyari Mohan Pradhan
10 September 2025 • AEU - International Journal of Electronics and Communications
With the increase in the number of users, the channel estimation in the presence of pilot contamination has become more challenging. To reduce the computational complexity involved in performing adaptive channel estimation in real time, block adaptive filters are widely used. This paper proposes a channel estimation technique that uses least mean square (LMS) adaptive filtering algorithm based on information-theoretic divergence, named Amari-Alpha divergence based Block LMS (AABLMS) algorithm. This algorithm is used to study a scenario where multiple users receive contaminated pilot signals. The condition for convergence of the proposed AABLMS algorithm in the mean sense is derived, and the upper and lower bounds for the learning rate are derived. Further, the block counterparts of existing state-of-the-art LMS variants are compared with that of the proposed AABLMS algorithm in terms of computational complexity, mean square deviation (MSD), and mean square error (MSE). The simulation results show that the proposed AABLMS algorithm performs better than other block LMS-based counterparts in the presence of channel noise and pilot contaminating noise.
Page updated
Google Sites
Report abuse