Sundararaja Sitharama Iyengar

Food fraud and adulteration continue to threaten food safety, consumer confidence and international trade. The challenges of investigating food-forensics reliably are compounded by increasingly complex supply chains, changing deceptive practices and difficulties in interpreting high-dimensional analytical data. The review critically evaluates current analytical chemistry and artificial intelligence (AI) approaches for the detection of adulteration, authenticity verification and determination of geographical or botanical origin in seafood, spices, meat, dairy products, oils, honey and beverages, distinguishing between direct forensic applications and complementary studies of food quality. The methodological framework includes liquid chromatography–mass spectrometry (LC-MS), headspace solid-phase microextraction coupled to gas chromatography–mass spectrometry (HS-SPME-GC-MS), isotope-ratio mass spectrometry (IRMS), Raman and near-infrared spectroscopy, and sensor arrays, chemometrics, machine learning (ML), convolutional neural networks (CNNs), explainable artificial intelligence (XAI), and multimodal data fusion. The reviewed literature shows that classical chemometric approaches are still useful for the small datasets usually employed in food-forensic studies, while deep learning offers more advantages for larger or high-dimensional datasets but the reliability is dependent on appropriate validation, model interpretability and chemically meaningful features. Future research should be geared toward developing standardized cross-laboratory datasets, strong external validation, explainable and data-efficient models, multimodal data fusion, portable analytical tools and regulatory structures to promote transparent and defensible forensic choices.

Federated Learning (FL) provides a flexible distributed platform where numerous clients with high data and system heterogeneity can collaborate to learn a model. While previous research has shown that FL can handle diverse data, it often completely assumes idealized conditions. In practice, real-world factors make it hard to predict or design individual client participation. This complexity results in an unknown participation pattern - arbitrary client participation (ACP). Hence, the key open problem is to understand the impact of client participation and develop a lightweight mechanism to support ACP in FL. In this paper, we first empirically investigate the client participation's influence in FL, revealing that FL algorithms are adversely impacted by ACP. To alleviate the impact, we propose a lightweight solution, Federated Average with Snapshot (FAST), that supports almost ACP for FL and can seamlessly integrate with other classic FL algorithms. Specifically, FAST enforces clients to take a snapshot once in a while and facilitates ACP for the majority of training processes. We prove that the convergence rates of FAST in non-convex and strongly-convex cases match those under ideal client participation. Furthermore, we empirically introduce an adaptive strategy to dynamically configure the snapshot frequency, tailored to accommodate diverse FL systems. Extensive experiments show that FAST significantly improves performance under ACP and high data heterogeneity.

The rapid development of mobile sensing technologies (e.g., smart devices embedded with various powerful sensors) has encouraged the proliferation of the Internet of Things (IoT). Although data reliability and accuracy are crucial in many sensor applications (e.g., air-quality monitoring), it is often difficult to ensure these properties. Mobile IoT’s people-centric architecture allows for more inaccurate and corrupted data. In this manuscript, we are addressing the problem of how to predict data more accurately in the presence of malicious participants who inject false data to manipulate the system. Our goal is to recover those missing or imprecise data values from the correlated data streams. To do so, we propose a Reputation-Based Context-Aware Data-Fusion (RCoD) mechanism that is resilient against on–off and data-corruption attacks. Furthermore, the Contextual Hidden Markov Model-based data prediction facilitates more accurate real-time data prediction. We tested the scenarios where most participants were malicious, injecting false data at varied rates. Our method accurately identified the honest participants based on their reported data and context. We empirically evaluate the performance using Beijing’s air-quality dataset. We compared the performance of our RCoD method against four state-of-the-art methods, and the results justify its superiority.

The exponential growth of the Internet of Things (IoT) has introduced considerable security and forensic challenges due to the rising complexity and heterogeneity of connected devices. As the adoption of the Internet of Things (IoT) continues to expand, so too do the vulnerabilities inherent to this technology, with threats ranging from the exploitation of individual devices to large-scale breaches of network security. It is imperative that IoT security frameworks undergo continuous evolution to address weaknesses in cryptography, authentication, and communication protocols. Concurrently, the field of IoT forensics encounters obstacles in the gathering and analysis of evidence due to the restricted memory and heterogeneous architectures of IoT devices. This chapter examines the critical aspects of IoT security, highlighting prevalent attacks, mitigation techniques, and the forensic methodologies employed to investigate compromised devices. Particular attention is given to the role of device heterogeneity, emerging forensic tools, and the impact of artificial intelligence on both security and forensic efforts. The discussion underscores the necessity for ongoing advancements to create a resilient IoT ecosystem capable of mitigating threats and enhancing forensic investigations.

The rapid growth of digital communication and video transmission has brought about an urgent need for robust encryption methods to ensure the security and integrity of video data. Traditional encryption techniques, while effective, face increasing challenges from advancing computing capabilities. Quantum cryptography has emerged as a promising solution to secure video transmission. In this article, we propose a novel cutting-edge Hybrid Quantum Video Encryption Framework. Our key idea is to integrate the strengths of quantum encryption and the classical video transmission paradigm to enhance the security of video transmission. By combining the proposed quantum encryption framework which uses a pseudorandom number generated key to perform a row-wise XOR operation along with the secure SSL-encrypted HTTP transmission over the internet, this framework provides a robust defense against threats. The experimental simulation conducted under the new computational integrated paradigm shows the strong encryption capability of the proposed method. The superior performance of the proposed method (ranges from 10-15% compared to the existing methods) as validated through statistical analysis, outperforms the state-of-the-art video encryption frameworks in terms of Information Entropy and Correlation Coefficient for encrypted plain images offering a significant advancement in the field of secure video transmission.

The advancement of IoT technologies influenced a substantial increase in the usage of multimedia devices. Consequently, forensic specialists began to experience a vast load of video and image data during the investigative procedures. This triggered a need to ensure the integrity of multimedia data and verify its source origin for digital forensics processes. In this paper, we address the problem of identifying the video source camera of the video data acquired by the investigators. We develop a novel convolutional neural networks (CNNs) ensemble framework to identify the video source camera. In our method, we analyze the video data using patches extracted from I-frames quadrants (i.e., non-overlapping squares) using independent CNNs for each quadrant to achieve location awareness. Experimental results demonstrate that our framework is robust for the same device type classification and outperforms existing deep learning-based techniques.