Pallav Kumar Deb

Pallav Kumar Deb

Auto-generated profile, based on publicly available information

Pallav Kumar Deb

Department of Computer Science and Engineering, Indian Institute of Technology Kharagpur, Kharagpur, India

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
22 December 2020 in IEEE Internet of Things Journal
Verified
Journal Article
Closed access
RSRituparna Saha
Rituparna Saha
SMSudip Misra
Sudip Misra
PDPallav Kumar Deb
Pallav Kumar Deb
In this paper, we propose a fog-enabled federated learning framework – FogFL – to facilitate distributed learning for delay-sensitive applications in resource-constrained IoT environments. While Federated Learning (FL) is a popular distributed learning approach, it suffers from communication overheads and high computational requirements. Moreover, global aggregation in FL relies on a centralized server, prone to malicious attacks, resulting in inefficient training models. We address these issues by introducing geospatially placed fog nodes into the FL framework as local aggregators. These fog nodes are responsible for defined demographics, which help share location-based information for applications with similar environments. Further, we formulate a greedy heuristic approach for selecting an optimal fog node for assuming a global aggregator’s role at each round of communication between the edge and cloud, thereby reducing the dependence on the execution at the centralized server. Fog nodes in the FogFL framework reduce communication latency and energy consumption of resource-constrained edge devices without affecting the global model’s convergence rate, thereby increasing the system’s reliability. Extensive deployment and experimental results corroborate that, in addition to a decrease in global aggregation rounds, FogFL reduces energy consumption and communication latency by 92% and 85%, respectively, as compared to state-of-the-art.

26 October 2022 in IEEE Transactions on Parallel and Distributed Systems
Verified
Journal Article
Closed access
RSRituparna Saha
Rituparna Saha
SMSudip Misra
Sudip Misra
ACAishwariya Chakraborty
Aishwariya Chakraborty
CCChandranath Chatterjee
Chandranath Chatterjee
PDPallav Kumar Deb
Pallav Kumar Deb
This work presents an efficient data-centric client selection approach, named DICE, to enable federated learning (FL) over distributed edge networks. Prior research focused on assessing the computation and communication ability of the client devices for selection in FL. On-device data quality, in terms of data volume and heterogeneity, across these distributed devices is largely overlooked. The obvious outcome is the selection of an improper subset of clients with poor-quallity data, which inevitably results in an inefficient trained model. With an aim to address this problem, in this work, we design DICE which prioritizes the data quality of the client devices in the selection phase, in addition to their computation and communication abilities, to improve the accuracy of FL. Additionally, in DICE, we introduce the assistance of vicinal edge devices to account for the lack of computation or communication abilities in certain devices without violating the privacy-preserving guarantees of FL. Towards this aim, we propose a scheme to decide the optimal edge device, in terms of latency and workload, to be selected as the helper device. The experimental results show that DICE improves convergence speed for a given level of model accuracy. Further, the simulation results show that DICE reduces delay by at least 16%, energy consumption by at least 17%, and packet loss by at least 55% compared to the existing benchmarks while prioritizing the on-device data quality across clients.

21 May 2021 in IEEE Internet of Things Journal
Verified
Journal Article
Closed access
NPNidhi Pathak
Nidhi Pathak
PDPallav Kumar Deb
Pallav Kumar Deb
AMAnandarup Mukherjee
Anandarup Mukherjee
SMSudip Misra
Sudip Misra
The atmospheric buoyancy and intangible nature of fatal communicable viruses lead to rapid transmissions among individuals, resulting in global pandemics. Strategic lockdowns and mandatory social distancing are immediate solutions in such scenarios. However, this leads to operational disruptions in education, manufacturing, economy, transportation, governance, and community. Although technological assistance is beneficial in overcoming such issues, the current Internet of Things (IoT) infrastructure has limitations. In this paper, we provide a comprehensive review of the possible IoT-based solutions that have the capacity of combating the COVID-19-like viruses. We highlight the societal impacts due to pandemics and identify the specific lacunae in current IoT solutions. We also provide comprehensive detail on how to overcome the challenges along with directions towards the possible technological trends for future research. Compared to existing reviews, our work offers a holistic view of the cause, effects, and the possible solutions that are existing, along with already existing solutions that can be customized to serve the special needs during the pandemic.

31 December 2020 in IEEE Internet of Things Journal
Verified
Journal Article
Closed access
SMSudip Misra
Sudip Misra
SRSri Pramodh Rachuri
Sri Pramodh Rachuri
PDPallav Kumar Deb
Pallav Kumar Deb
AMAnandarup Mukherjee
Anandarup Mukherjee
The Internet of Things (IoT) environments have hard real-time tasks that need execution within fixed deadlines. As IoT devices consist of a myriad of sensors, each task is composed of multiple interdependent subtasks. Towards this, the cloud and fog computing platforms have the potential of facilitating these IoT Sensor Nodes (SNs) in accommodating complex operations with minimum delay. To further reduce operational latencies, we breakdown the high-level tasks into smaller subtasks and form a Directed Acyclic Task Graph (DATG). Initially, the SNs offload their tasks to a nearby Fog Node (FN) based on a greedy choice. The greedy formulation helps in selecting the FN in linear time while avoiding combinatorial optimizations at the SN, which saves time as well as energy. IoT environments are highly dynamic, which mandates the need for adaptive solutions. At the chosen FN, depending on the dependencies on the DATGs, its corresponding deadlines, and the varying conditions of the other FNs, we propose an -greedy non-stationary multiarmed bandit-based scheme (D2CIT) for online task allocation among them. The online learning D2CIT scheme allows the FN to autonomously select a set of FNs for distributing the subtasks among themselves and executes the subtasks in parallel with minimum latency, energy, and resource usage. Simulation results show that D2CIT offers a reduction in latency by 17% compared to traditional fog computing schemes. Additionally, upon comparison with existing online learning-based task offloading solutions in fog environments, D2CIT offers an improved speedup of 59% due to the induced parallelism.

28 December 2022 in IEEE Communications Surveys & Tutorials
Verified
Journal Article
Closed access
RARahul Ajmeria
Rahul Ajmeria
MMMayukh Mondal
Mayukh Mondal
RBReya Banerjee
Reya Banerjee
THTamesh Halder
Tamesh Halder
PDPallav Kumar Deb
Pallav Kumar Deb
DCDebashish Chakravarty
Debashish Chakravarty
Industrial Internet of Things (IIoT) and its applications have seen a paradigm shift since the advent of artificial intelligence and machine learning. However, these methods are mostly data-centric and lack flexibility. Brain-Computer Interface (BCI) is envisioned as an efficient solution that may overcome the shortcomings of data-centric techniques by introducing human intuition in the process loop. In this work, we present an exhaustive study on the feasibility of adopting BCI techniques for industrial applications, particularly Electroencephalography (EEG). We present a comprehensive literature survey on the basics of EEG (including signal processing techniques) and its involved paradigms, together with its application scope. We identify the potential use cases of EEG-based BCI systems in industries in the current traction while accounting for their pros and cons. Additionally, we also highlight the existing challenges and propose possible off-the-shelf solutions for overcoming them. We present a two-pronged study (hardware and software specifications) on its deployment by performing lab-scale experiments with a single-channel EEG headset. We compare the metrics of our experiments to 64, 32, and 25 channel data from other studies (collected from the Internet). On average, we observe CPU usage of up to 35%, memory usage of 1.75 Gb, and data transfer rates of 25 Mbps. In contrast to the conventional setups, we demonstrate how a minimalistic setup using a single-channel EEG headset may facilitate complex applications like job inspection in manufacturing processes, with an overall accuracy of 70%.

4 June 2021 in IEEE Transactions on Network Science and Engineering
Verified
Journal Article
Closed access
PDPallav Kumar Deb
Pallav Kumar Deb
AMAnandarup Mukherjee
Anandarup Mukherjee
SMSudip Misra
Sudip Misra
In this paper, we propose a delay-aware Q-learning-based routing algorithm - XiA - for sending data from users to nearest Access Points (APs) through Unmanned Aerial Vehicle (UAV) swarms communicating using 6G technology. These UAVs assist the ground networks in overcoming communication voids while maneuvering through different demographics. However, the communication links in the THz band have limited transmission range, causing the UAVs to frequently disconnect from the swarm. We overcome such issues by waiting until the UAV comes in contact with others in case of non-time-sensitive data. In the case of time-sensitive data, the UAVs send the data to the APs through Low Earth Orbit (LEO) satellites. To empower XiA to adapt to the changing environments and expensive delays in LEO, we model the rewards by accounting for spreading and absorption in the 6G channels, and Doppler effect and pointing error in the satellite channel. We show our bias for the parameters through extensive simulations and prove that the Q-model in XiA achieves convergence under all conditions. Additionally, in comparison with state-of-the-art solutions, we observe that XiA offers an improved delay of 82%.