J. Harshan

Soumita Hazra, J. Harshan

Motivated by recent developments in full-duplex radios, cognitive radios, and data-driven signal-processing, we propose a novel class of reactive jamming adversaries wherein the adversary transmits jamming energy on the victim's frequency band while simultaneously monitoring various energy statistics in the network to detect the presence of potential countermeasures, thereby trapping the victim. These adversaries employ generalized energy detectors comprising statistical detectors, based on instantaneous and distributional energy metrics, and data-driven detectors employing machine-learning classifiers to learn patterns in the observed energy sequences. Against such a strong adversary, we propose a family of cooperative mitigation strategies wherein the victim takes assistance from a helper node, with the strategies tailored to operate under a wide range of latency requirements on victim's messages and practical radio hardware constraints at helper node. To provide theoretical guarantees on their efficacy, interesting optimization problems are formulated on the choice of their underlying parameters, followed by extensive mathematical analyses on their error performance and covertness. To assess their practical feasibility, we implement the before-deployment and after-deployment setups on a software-defined-radio-based hardware testbed, and to evaluate their detectability on real energy observations, we collect the corresponding datasets to train and test the data-driven machine-learning classifiers employed by adversary. Both analytical and hardware evaluations show that the proposed strategies cannot be detected with a high-probability under the generalized energy detectors used by adversary.

Arpit Guleria, J. Harshan, Ranjitha Prasad et al.

Class imbalance in training datasets can lead to bias and poor generalization in machine learning models. While pre-processing of training datasets can efficiently address both these issues in centralized learning environments, it is challenging to detect and address these issues in a distributed learning environment such as federated learning. In this paper, we propose FLICKER, a privacy preserving framework to address issues related to global class imbalance in federated learning. At the heart of our contribution lies the popular CKKS homomorphic encryption scheme, which is used by the clients to privately share their data attributes, and subsequently balance their datasets before implementing the FL scheme. Extensive experimental results show that our proposed method significantly improves the FL accuracy numbers when used along with popular datasets and relevant baselines.