With rapid technological growth, security attacks are drastically increasing. In many crucial Internet-of-Things (IoT) applications such as healthcare and defense, the early detection of security attacks plays a significant role in protecting huge resources. An intrusion detection system is used to address this problem. The signature-based approaches fail to detect zero-day attacks. So anomaly-based detection particularly AI tools, are becoming popular. In addition, the imbalanced dataset leads to biased results. In Machine Learning (ML) models, the $F_{1}$ score is an important metric to measure the accuracy of class-level correct predictions. The model may fail to detect the target samples if the $F_{1}$ is considerably low. It will lead to unrecoverable consequences in sensitive applications such as healthcare and defense. So, any improvement in the $F_{1}$ score has a significant impact on the resource protection. In this paper, we present a framework for an ML-based intrusion detection system for an imbalanced dataset. In this study, the most recent dataset, namely CICIoT2023 is considered. The random forest (RF) algorithm is used in the proposed framework. The proposed approach improves 3.72%, 3.75% and 4.69% in precision, recall and $F_{1}$ score, respectively, with the existing method. Additionally, for unsaturated classes (i.e., classes with $F_{1}$ score < 0.99), $F_{1}$ score improved significantly by 7.9%. As a result, the proposed approach is more suitable for IoT security applications for efficient detection of intrusion and is useful in further studies.
Research on major depressive disorder (MDD) is one of the major fields of world health. Among all the solutions for depression identification, electroencephalography (EEG) is a very useful tool that has received a broad range of attention from researchers. In this study, we have proposed a novel deep learning method based on the spatial-temporal graph convolutional network (ST-GCN) in combination with depression-related functional connectivity graphs. In this work, the differential entropy (DE) feature of EEG is obtained and the adjacency matrix of the brain graph is created using the Phase Locking Value (PLV) matrix between EEG signal pairings. A proportional threshold is trained to select the critical edges of the brain graph to eliminate weak associations. For the classifier, a combined prior knowledge ST-GCN network constructed by spatial convolutional blocks and standard temporal convolutional blocks is employed to improve the spatial-temporal feature learning ability. Compared with other methods, our method obtains the best accuracy of 93.85%. The sound performance demonstrates the potential of the ST-GCN combined with the depression-related functional connectivity graph for clinical diagnostic and treatment prediction.
Approximate computing is an emerging paradigm in error-tolerant applications that leads to power-efficient designs without significant loss in quality. The divider in these applications have complex hardware and more latency among the computational blocks resulting in power consumption. Hence approximating the division module would lead to designs with vastly improved power efficiency. A new approximate subtractor (AxSUB) is proposed in this paper with the intent to reduce the hardware complexity while achieving accuracy within permissible limits. The proposed AxSUB and existing approximate subtractor units are used in the restoring array division (RAD) architecture to prove the efficacy of the AxSUB. Comprehensive error and synthesis analysis are performed on RAD architectures implemented using AxSUB, and existing methods. Our proposed design achieved a 21% decrease in area and a 28% decrease in power consumption compared to the exact design. The proposed and existing RAD architectures is implemented on change detection applications to validate the quality-effort tradeoff.
Adaptive filter is a concept that is gaining interest in recent years as they start to find practical real-time applications in fields like channel equalisation, echo cancellation, noise cancellation, and a variety of other adaptive s ignal processing applications. This article proposes a VLSI architecture for an adaptive filter application with a control unit that combines noise reduction and echo cancellation techniques. For self-adjustment of the filter weights, the least mean square (LMS) algorithm is utilised. The VLSI implementation and performance analysis provide crucial information about the structure of an algorithm, such as hardware requirements, power consumption, and real-time performance. The performance of the proposed architecture of the adaptive filter is assessed in terms of delay, area, and power dissipation using $45\ nm$ CMOS technology.
Medical image watermarking (MIW) is a widely used method today to confirm the validity of medical data. This paper proposes a hybrid, robust, semi-blind and imperceptible MIW system using the Discrete Cosine System (DCT) and Fast Randomized Singular Value Decomposition (FRSVD) with less computational complexity and computational time; relevant to both medical and non-medical images. Multiple scans and assessment matrices are used to evaluate the efficacy of the suggested technique, and it is compared to cutting-edge methodologies. The procedure obtained a maximum Peak Signal to Noise Ratio (PSNR) of 81.71 dB, Structural Similarity Index (SSIM) of 1, and Normalized Coefficient (NC) value of 1 for all the attacks. The suggested scheme achieved a meager embedding time and extraction time; 0.1638 seconds and 0.0955 seconds, respectively.
