Kalyan Nakka

The growing adoption of Large Language Models (LLMs) has influenced the development of Small Language Model (SLMs) for on-device deployment across smartphones and edge devices, offering enhanced privacy, reduced latency, server-free functionality, and improved user experience. However, due to on-device resource constraints, SLMs undergo size optimization through compression techniques like quantization, which inadvertently introduce fairness, ethical and privacy risks. Critically, quantized SLMs may respond to harmful queries directly, without requiring adversarial manipulation, raising significant safety and trust concerns. To address this, we propose LiteLMGuard, an on-device guardrail that provides real-time, prompt-level defense for quantized SLMs. Additionally, our guardrail is designed to be model-agnostic such that it can be seamlessly integrated with any SLM, operating independently of underlying architectures. Our LiteLMGuard formalizes deep learning (DL)-based prompt filtering by leveraging semantic understanding to classify prompt answerability for SLMs. Built on our curated Answerable-or-Not dataset, LiteLMGuard employs ELECTRA as the candidate model with 97.75% answerability classification accuracy. The on-device deployment of LiteLMGuard enabled real-time offline filtering with over 85% defense-rate against harmful prompts (including jailbreak attacks), 94% filtering accuracy and  135 ms average latency. These results demonstrate LiteLMGuard as a lightweight robust defense mechanism for effectively and efficiently securing on-device SLMs against Open Knowledge Attacks.

Coverage and connectivity are the major performance metrics that reflect the quality of service provided by heterogeneous planar wireless sensor networks (PWSNs) monitoring a planar field of interest (PFoI), where the deployed sensors may not necessarily have the same capabilities in terms of their sensing range, communication range, and maximum battery power capacity. Precisely, coverage is considered as the main functionality of PWSNs, which is meaningful only when connectivity is guaranteed. Therefore, it is important that both coverage and connectivity be jointly considered in the deployment of heterogeneous PWSNs. In order to account for joint coverage and connectivity, we propose to solve the problem of connected k-coverage in heterogeneous PWSNs, where every point in a PFoI is covered by at least k sensors simultaneously, while all the deployed sensors are mutually connected, either directly or indirectly, with k > 1. While most existing studies of this problem focus on homogeneous sensors, which have the same above-mentioned capabilities (i.e., initial energy, sensing range, and communication range), our study in this paper considers heterogeneous ones. More specifically, we propose a hierarchical (or multi-tier) deployment of heterogeneous sensors in a square FoI, which is divided into concentric square bands with the same width difference to achieve k-coverage of this PFoI. Based on this multi-tier sensor deployment and the slicing of a square FoI into square bands for k-coverage, we establish the necessary relationship for connectivity among the sensors located in adjacent bands. Finally, we propose our heterogeneous k-coverage protocol and validate our theoretical analysis using simulation results. We find that the deployment of heterogeneous sensors helps achieve much better results compared to those obtained using homogeneous sensors. Furthermore, our proposed protocol outperforms an existing connected k-coverage protocol for heterogeneous PWSNs with respect to various performance metrics.

In the design of planar wireless sensor networks (PWSNs), the preliminary tasks are to achieve both coverage and connectivity, which are essential for the correct operation of this type of network. A PWSN is said to be in perfect operational condition only when it is capable of guaranteeing both coverage of the field and connectivity among all the active sensor nodes. In order to achieve both coverage and connectivity in PWSNs, we intend to solve the problem of connected k-coverage in PWSNs, where each point in a field of interest is covered (or sensed) by at least k sensor nodes (k > 1) simultaneously and all the participating sensor nodes in the k-coverage process are connected to each other. In our study, we found an irregular hexagon, denoted by IrHx(rs/n), as the best polygon for tessellating a field of interest and allowing to deploy sensor nodes, where n > 1 is a natural number and rs is the radius of the sensing range of the sensor nodes. First, we construct our proposed irregular hexagon, IrHx(rs/n), which forms a tile, using the regular hexagon. Second, we compute the minimum sensor density required to k-cover a planar field of interest using our proposed IrHx(rs/n) tile. Third, we establish a relationship between the sensing and communication radii of the sensor nodes to ensure network connectivity in k-covered PWSNs. Finally, we substantiate our theory with various simulation results.

Quantum era is near and advanced quantum computing will pose a huge threat to the security of the current cryptographic systems in distributed energy resources (DER)rich power grids. It is necessary to prepare now to combat quantum computing attacks with serious real-world consequences in the years ahead. Recently, post-quantum cryptography (PQC) is considered as one of the major candidates of quantum attack defense strategies, and the adoption of PQC for DER systems has not been fully studied yet. This paper discusses the adoption of PQC in a standard DER network protocol (i.e., IEEE 2030.5-PQC), and proposes a real-time hardware-in-the-loop co-simulation PQC testbed consisting of a DER physical system simulation and a cyber system simulation such as DER gateways and DER management system (DERMS) server. Universal custom-made PQC client and server software are developed to meet the compliant with IEEE 2030.5-2018 standard and implemented in the DER gateways and the DERMS server, respectively. Finally, the feasibility of the proposed PQC-grade DER network is validated by using the real-time co-simulation testbed.