The latest A2A systems often use LLM-based agents for tool calling and planning workflows, introducing risks such as prompt injection attacks. We introduce a lightweight, proof-carrying receipt for agent-to-agent (A2A) messaging that supports fully offline verification. This receipt enables third parties to verify who authored a message and confirm whether it was logged, without contacting the sender, the receiver, or any online service. An agent signs a canonical JSON representation of each message. Servers log the payload hash in a Merkle tree and periodically sign the tree state as a Signed Tree Head (STH). The returned bundle includes a receipt containing a Merkle proof, an STH, and a minimal decentralized identifier (DID) registry for offline key discovery. A verifier recomputes the Merkle root and validates the payload-hash match, Merkle inclusion, and both the STH and client signatures entirely offline. In our prototype, the proof size grows as $O(\log N)$, and the median offline verification time is approximately 0.4 s on our test platform. Receipts self-authenticate individual messages within a single transparency log, providing a lightweight alternative to other approaches.
Traditional data anonymization techniques-including generalization, suppression, perturbation, and masking-were designed for an era of limited contextual modeling. These methods effectively balanced privacy protection and data utility for structured, textual, audio, and visual data. However, the emergence of Large Language Models (LLMs) and multimodal foundation architectures has fundamentally disrupted this balance. Modern models, equipped with powerful representation-learning and cross-modal reasoning, can infer or reconstruct masked identities using contextual, semantic, or visual cues, rendering traditional anonymization insufficient. This paper surveys anonymization techniques across tabular, textual, audio, and video domains, comparing their effectiveness before and after the advent of LLMs. Through controlled experiments, we demonstrate that context-aware LLMs can accurately re-identify subjects from anonymized datasets, even when explicit identifiers are removed. The results reveal that traditional anonymization fails to address contextual inference threats, calling for a paradigm shift toward holistic, multi-layered, privacy-preserving approaches. We discuss emerging strategies-such as differential privacy, federated learning, synthetic data generation, and LLM-in-the-loop anonymization-as potential solutions for privacy preservation in the LLM era.
This paper introduces DP-Prox, an end-to-end framework for robust and differentially private federated instruction tuning of small Large Language Models (LLMs) on 8 GB edge devices. DP-Prox is instantiated using the 3.8 billionparameter Phi-3-mini model with 4-bit Quantized Low-Rank Adaptation (QLoRA), and combines (i) FedProx-style proximal regularization on low-rank adapters to stabilize optimization under highly Non-IID smart-home data, and (ii) Differentially Private SGD (DP-SGD) applied directly to adapter gradients to provide user-level $(\epsilon, \delta)$-differential privacy against an honest-but-curious server. In a realistic IoT instruction-following benchmark with Dirichlet-skewed clients ($\alpha=0.3$), DP-Prox improves F1-score from 0.80 (Edge-FIT with FedAvg) to 0.82 while reducing convergence rounds from $T=50$ to $T \approx 40$. Under a strict privacy budget $\left(\epsilon=5, \delta=10^{-5}\right)$, DP-Prox maintains an F1-score of 0.77 with only $+15 \%-20 \%$ local overhead, and preserves the communication efficiency of QLoRA-based federated training. Qualitative analyses further show that DP-Prox yields more consistent instruction-following behavior across heterogeneous homes, especially on multi-device routines and privacy-sensitive queries. Together, these results position DP-Prox as a practical and regulation-ready blueprint for deploying small, edge-resident LLMs in smart-home IoT environments.
Large Language Models (LLMs) have transformed many areas of artificial intelligence (AI), but their substantial resource requirements limit their deployment on mobile and edge devices. Although LLMs have demonstrated impressive capabilities in various applications, the next challenge is deploying these models on resource-limited edge devices. Recent innovations in software optimizations and specialized hardware have made it feasible to deploy LLMs on edge devices. Edge deployment offers benefits including local data processing, cost savings, reduced latency, and personalized responses. This paper reviews the latest techniques that make the deployment of LLMs on edge devices feasible and provides actionable, constraint-driven guidelines for practical edge deployment.
Ahstract- The healthcare sector faces significant challenges in balancing adaptability, security, and resource efficiency in medi-cal file management. This paper introduces a lightweight protocol that optimizes cryptographic components for using ephemeral key encapsulation, combining adaptive Challenge Response Pair with robust error correction. Designed for limited resource environments, the proposed approach minimizes computational overhead, ensures reliable key recovery, and provides privacy protection. The execution results validate its performance, making it a practical solution for a decentralized healthcare system.
New case breast cancer has steadily increased since 2012 and contributes to highest treatment costs among cancers. While current literature widely applies Machine learning (ML) and Artificial Intelligence (AI) in breast cancer prediction using patient features. However, latent relationships between diseases and patients remain unexplored. This study utilizes National Health Interview Survey (NHIS) 2023 and focused on chronic diseases participants. 738 breast cancer patients and 733 non-breast cancer patients are randomly selected. Patient-centric network attributes and patient features are both included into seven machine-learning risk prediction models. XGBoost model outperform among others having an Area Under Curve (AUC) of 94%. Notably, eigenvalue centrality and obesity are the most important features. This novel approach represents a significant improvement over current literature and provides promising applications in disease prediction.
As blockchain technology and smart contracts become widely adopted, securing them throughout every stage of the transaction process is essential. The concern of improved security for smart contracts is to find and detect vulnerabilities using classical Machine Learning (ML) models and fine-tuned Large Language Models (LLM). The robustness of such work rests on a labeled smart contract dataset that includes annotated vulnerabilities on which several LLMs alongside various traditional machine learning algorithms such as DistilBERT model is trained and tested. We train and test machine learning algorithms to classify smart contract codes according to vulnerability types in order to compare model performance. Having fine-tuned the LLMs specifically for smart contract code classification should help in getting better results when detecting several types of well-known vulnerabilities, such as Reentrancy, Integer Overflow, Timestamp Dependency and Dangerous Delegatecall. From our initial experimental results, it can be seen that our fine-tuned LLM surpasses the accuracy of any other model by achieving an accuracy of over 90%, and this advances the existing vulnerability detection benchmarks. Such performance provides a great deal of evidence for LLMs' ability to describe the subtle patterns in the code that traditional ML models could miss. Thus, we compared each of the ML and LLM models to give a good overview of each model's strengths, from which we can choose the most effective one for real-world applications in smart contract security. Our research combines machine learning and large language models to provide a rich and interpretable framework for detecting different smart contract vulnerabilities, which lays a foundation for a more secure blockchain ecosystem.
The effective use of artificial intelligence (AI) to enhance cyber security has been demonstrated in various areas, including cyber threat assessments, cyber security awareness, and compliance. AI also provides mechanisms to write cybersecurity training, plans, policies, and procedures. However, when it comes to cyber security risk assessment and cyber insurance, it is very complicated to manage and measure. Cybersecurity professionals need to have a thorough understanding of cybersecurity risk factors and assessment techniques. For this reason, artificial intelligence (AI) can be an effective tool for producing a more thorough and comprehensive analysis. This study focuses on the effectiveness of AI-driven mechanisms in enhancing the complete cyber security insurance life cycle by examining and implementing a demonstration of how AI can aid in cybersecurity resilience.
The paper outlines a thorough Application Specific Integrated Circuit (ASIC) design flow for Universal Asynchronous Receiver and Transmitter (UART) utilizing RTL to GDS implementation and verification. Hardware verification employs the Universal Verification Methodology (UVM), while the OpenROAD tool is integrated for physical designs spanning synthesis to Graphic Data System (GDS). The paper states the relevance of UART in real-time systems, particularly in the context of the increasing prevalence of IoT and Big Data applications. It delves into the encountered challenges during the research project and offers recommendations for future advancements. Through our experimentations with the OpenROAD, we demonstrated the design and verification process of the sophisticated digital circuit. The results of the verification are shown via waveforms for test cases. The results of the backend implementation include layouts and a QoR (Quality of Results) report table. The paper concludes by stating the significance of each stage in the ASIC design cycle with future scope in optimizing the design for heightened frequencies and enhanced performance.
Mental health challenges are on the rise in our modern society, and the imperative to address mental disorders, especially regarding anxiety, depression, and suicidal thoughts, underscores the need for effective interventions. This paper delves into the application of recent advancements in pretrained contextualized language models to introduce MindGuide, an innovative chatbot serving as a mental health assistant for individuals seeking guidance and support in these critical areas. MindGuide leverages the capabilities of LangChain and its ChatModels, specifically ChatOpenAI, as the bedrock of its reasoning engine. The system incorporates key features such as LangChain's ChatPrompt Template, HumanMessage Prompt Template, ConversationBufferMemory, and LLMChain, creating an advanced solution for early detection and comprehensive support within the field of mental health. Additionally, the paper discusses the implementation of Streamlit to enhance the user experience and interaction with the chatbot. This novel approach holds great promise for proactive mental health intervention and assistance.
Vehicular traffic and congestion is a major challenge worldwide because of rapid growth in urban population. The congestion can be mitigated to enhance traffic management by predicting accurate travel time of the vehicles in the traffic. This research developed a novel methodology utilizing machine learning on real-time traffic data collected through Bluetooth sensors deployed at traffic intersections to estimate travel time. The research evaluates performance and accuracy of five different prediction systems for travel time estimation highlighting the effectiveness of the machine learning models in accurately predicting travel time. The research also explores the development of the machine learning model predicting the travel time during peak hours, considering traffic lights impact on travel time between intersections. This research findings contribute to the efficient and reliable travel time prediction systems development, helping commuters making informed decisions and improve traffic management strategies.
A two stage algorithm developed called Multiple Gain Adaptations for Improved Networks (MGAIN) is presented. MAGAIN alternatively finds output weights and uses several gain factors to update the input weights in a Multi-Layer Perceptron. The gain factors are computed using Newtons method. Our method dynamically adjusts the quantity of gain factors calculated to maximize the reduction in loss with each epoch. The results demonstrate that our approach outperforms existing second order algorithms across the majority of diverse datasets.
With the rapid development of modern electronics and computation capability, biomechanical mining is attracting more and more attention. Due to the complexity and inter-person variance among the biomechanical dynamics, it is important to study the spatial variability to not only optimize the activity recognition performance but also deepen the understanding of inter-sensor-location and inter-subject differences. We here propose a system with both wearable motion sensing and deep learning, for comprehensive biomechanical dynamics understanding. More specifically, the system allows collection of motion data from diverse body locations. Further, the deep learning algorithm then learns the signals and yields the physical activity types. The experiments have promisingly indicated the spatial variability of biomechanical dynamics capturing and analysis. This study will benefit the understanding of biomechanical dynamics.
Digital signal processing (DSP) applications are known for their tolerance to the existence of minor errors in their output. This error tolerance feature leads to interesting optimizations of the hardware co-processors of such applications using approximate computing. In particular, approximate multiplication - a major component in DSP hardware- is one of the most trending topics nowadays since it increases the power efficiency and reduces the space and time complexities for applications that are error flexible. This paper investigates the existing multiplier design based on the proposed 6 different approximate 8-bit multiplier designs based on four approximate 4-2 compressors (three existing and one new), an approximate full adder, and an existing approximate half adder blocks. The accuracy of the proposed approximate multipliers is demonstrated through the image multiplication of two sample images. In addition, the performance of the approximate multipliers is demonstrated in terms of area, speed, and power consumption based on Application specific integrated circuits (ASIC) realizations using the open source NANGate 45nm CMOS libraries. Significant increase in performance can be observed when approximate compressor are used. Approximate multiplier AFA App. Mul4 brings around 13% reduction in area, 14.5% reduction in power whereas the highest speed is achieved by the approximate multiplier App. Mul5 brings down the delay by around 3.6% when compared with counterparts.
The impact of the spread of Harmful Algal Blooms (HABs) in water bodies has affected fisheries, public health, and tourism. This necessitates the need for an effective watershed monitoring framework for HABs in lakes. Several factors play a role in the formation of HABs and thus require careful evaluation of real-time data. Currently, many of such data are stored as silos on different databases in different monitoring platforms. This paper focuses on developing a conceptual framework by unifying multi-source data for monitoring HABs in lakes. The framework was designed to focus on ensuring availability as the quality attribute. Based on the logical view representation of the framework, availability was guaranteed through the concept of sensing.
We describe an extension to margin-based linear models for multiclass classification in which the cost function is learned alongside the parameters. The intuition is to exploit estimates of the relative "ease" of avoiding different types of errors, allowing the learner to focus on easier distinctions while "giving up" on harder ones.
Wireless Sensor Networks (WSNs) are becoming more essential in areas such as environmental monitoring, industrial automation, and healthcare. However, their wireless nature leaves them vulnerable to security threats such as eavesdropping, tampering, and attacks. This paper introduces a robust and secure algorithm for WSNs that enhances their security and reliability by incorporating key distribution, packet timestamping, location-based detection, wormhole detection, packet encryption, and wormhole prevention. Our algorithm utilizes the TwoFish algorithm to establish a shared key, timestamps each data packet using a synchronized network time protocol, monitors hop count, determines the distance between nodes using RSSI or other location-based techniques, exchanges probe packets to detect wormhole attacks, and encrypts data packets with TwoFish. In the event of a wormhole assault, it isolates the affected network segment by obstructing communication in that area. The efficacy of this approach is validated through implementation in a simulated NS-2 network environment, demonstrating significant enhancements in the security and reliability of WSNs against wormhole attacks.
This paper designs a surface electromyography (sEMG) amplification system capturing high density sEMG (HD-sEMG) with a maximum sampling frequency between 300 and 500 Hz. The system has been designed in ordinance with current IEEE research on sensor-electrode characteristics and is intended as a launchpad for future open source sEMG research. This paper does not implement the designed wristband but serves as a literature review and open-source design using commercially available technologies.
The latest developments in low-power wireless communication systems like Wi-Fi and LoRa have enabled new capabilities in large-scale Wireless Sensor Networks (WSN), specifically in agricultural monitoring and environmental monitoring applications. The long-range transmission and the adaptable ecosystem have enforced several challenges to maintain these WSNs. Temperature plays a crucial role in the reliability of the electronics, as temperature increases, the failure of the device increases exponentially. In this paper, a methodology to monitor the health of the node periodically along with the ability to identify an unhealthy node is proposed which eases the maintenance of the WSN. There is no existing methodology that addresses this issue at the nodal level. This proposed novel methodology solves the problem of monitoring and maintaining the health of the node of a WSN.
This paper presents the ways edge computing can influence both application performance and client-side battery consumption by testing a standard web application with varying configurations. Through these tests it was possible to determine that edge computing can have a measurable effect on mobile application battery consumption and may be able to increase the longevity of client device batteries if taken into consideration when designing an application. There are also some interesting results that indicate a relational database may outperform a cloud-native NoSQL database under certain conditions, but more testing and analysis will be required to determine the basis of the results and validate the findings in this area.
