Sudhakar Kumar

Air quality plays a critical role in human well-being, as air pollution significantly contributes to respiratory diseases such as pneumonia, cysts, and asthma. Predicting pollution levels can enable targeted interventions to mitigate associated chronic health risks. According to the World Health Organization, outdoor air pollution is responsible for approximately 4.2 million premature deaths worldwide. This paper proposes a PPDRTL framework to predict roadside air quality based on pollutant deposition on tree leaves. High-resolution images of roadside leaves are captured and analyzed to estimate pollution levels at the street scale. Image features, including contrast, entropy, and standard deviation, are extracted under varying traffic conditions, namely high, medium, and low traffic densities. Segmentation techniques such as PSO, DPSO, and FODPSO are employed to enhance pollutant feature extraction from leaf surfaces. The PPDRTL framework utilizes a Linear Regression model to predict air quality index (AQI) values from leaf image features, while ground truth data are obtained from the Haryana State Pollution Control Board (HSPCB) for validation. For the 10-day pilot dataset, FODPSO-based PPDRTL achieves R ² up to 0.894, with RMSE in the range of 8.76 − 12.34   μ g / m 3 across high-, medium-, and low-traffic sites. Furthermore, the framework achieves prediction accuracies of 88.91% for high-traffic areas, 90.70% for medium-traffic areas, and 91.42% for low-traffic areas, demonstrating its effectiveness as a robust approach for fine-grained air quality prediction and environmental monitoring.

As modern system-on-chip platforms become more complex, they are more exposed to advanced runtime threats that traditional security methods have difficulty handling. Threats like hardware Trojans, malicious firmware, microarchitecture side channels, and physical attacks are increasingly common on today's SoC platforms. Traditional design-time verification and broad monitoring are not enough to detect changing adversarial behaviors. Therefore, ensuring security during real-time system operation is now a key challenge. An extensive overview of on-chip intelligence, a paradigm in which lightweight embedded machine-learning engines continuously profile system activity and identify anomalies at runtime, is provided in this chapter. Modern methods for identifying power-timing manipulations, Trojan horses, side-channel anomalies, and fault injection attempts are assessed. Additionally, we present an architectural blueprint for next-generation self-defending chips that incorporate cross-layer monitoring fabrics, secure enclaves, trace buffers, and AI accelerators.

Authentic, trusted semiconductor components are essential for safety-critical systems, but traditional manual, documentation-based verification cannot keep pace with complex, globalized chip supply chains and increasingly sophisticated counterfeit, recycled, and remarked ICs. AI-based technologies address this gap by detecting subtle counterfeit indicators and hidden patterns in supply chain data, focusing particularly on cases where conventional methods fail to flag suspicious parts. In this context, AI-encoded systems support semiconductor authenticity through three main capabilities: vision-based inspection for chip and packaging features, AI-generated data for origin and provenance verification, and AI-driven evaluation for smart trust scoring across suppliers and lots. These approaches enable automated chip inspection and anomaly detection in supply chain behavior, and their performance can be directly compared with traditional non-AI verification to demonstrate gains in accuracy, coverage, and reliability.

The rising prevalence of substance abuse and overdose incidents underscores the need for real-time public health surveillance. Social media offers valuable signals for monitoring these events; however, noisy language, slang usage, and class imbalance present significant challenges for automated analysis. To address these issues, the authors propose ATTEND, a multi-task neural network for substance classification and detection of 18 overdose symptoms, with symptom normalization to standardized MedDRA concepts. ATTEND was trained on a large multi-source corpus combining ADE Corpus V2 and the UCI Drug Review Dataset, comprising over 100,000 samples designed to emulate realistic social media communication. Experimental results show that ATTEND achieved 93.23% accuracy and 93.41% weighted-F1 for substance classification, 94.10% micro-F1 for overdose symptom detection, and 90.42% accuracy for symptom normalization, outperforming baseline multi-task models across all tasks. The framework is scalable, privacy-preserving, and suitable for real-time monitoring of drug abuse signals.

Pattern recognition is key to enhancing automated decision systems, particularly for image data analysis. As visual data grows in volume and complexity, robust models for accurate pattern interpretation are essential. This study examines multi-class image recognition via advanced architectures including CNNs, hybrid CNN-RNNs, and Vision Transformers (ViT), with comparative experiments. Performance metrics include accuracy, precision, recall, and F-score. The hybrid CNN-RNN model excelled, achieving 92.24% test accuracy by fusing spatial and temporal features. Precision and recall exceeded 90%, proving effectiveness across patterns. This work demonstrates how advanced methods boost classification accuracy, aiding smarter AI for real-world tasks.