Search

Detection of anomalies in multivariate time series (MVTS) data has been crucial for various application areas. Though recent approaches have addressed such issues, still the problem is challenging not only due to identifying anomalous behavior but also to reducing false alarms. The problem becomes more exciting and challenging when the model takes care of errors due to measurement, environmental effects, and temporal dependency among the observations. Toward this, we propose a framework MAD-MEL, an MVTS Anomaly Detection Method using metric and entity learning (EL) approaches. The metric learning-based model identifies the behavior of individual series as well as the relationship between different attributes, whereas the entity-based model learns the behavior of MVTS as a single entity. An empirical mode decomposition-based feature learning (FL) component is incorporated to learn more generalized behavior of data and to handle the problem of minor disturbances. Also, the proposed method combines a linear component to control the spatial scaling effect in the train and test set. The idea is to learn the normal behavior of MVTS by combining all the components as a prediction-based model and to use prediction error for anomaly detection. We adopt transfer learning to reduce the time complexity while learning the individual series in case of a large number of attributes or multiple instances of MVTS. Our proposed multivariate time series anomaly detection using metric and entity learning (MAD-MEL) achieves the F1-score of 94% on public dataset, which is an improvement of 3% from SOTA.

Recommendations generated by a model become more convincing when the system is capable of explaining the rationale behind the recommendations with respect to various decision parameters involved. A recommendation system which uses fuzzy expert system and self attention (FESSARec) to explain the recommendations is proposed here. The self-attention module extracts features of learners and courses and generates attention weights which will be used to explain the recommendations. The fuzzy expert system extracts relevant rules from the additional domain knowledge available in the datasets. As a result of this hybrid approach, FESSARec outperforms the recent architectures with which it is compared and obtains a very small root mean square error (RMSE) score of 0.65. FESSARec is also capable of producing top-N recommendations with a very high NDCG of 0.89 and HR of 0.72. It outperforms the best e-commerce baseline by 8% and the educational baseline by 16% of lower error rates.

Intrinsic manufacturing process variations are extensive, unpredictable, and inevitable in modern semiconductor technology. This is advantageously utilized in physically unclonable function (PUF) circuits for enhancing hardware and software security. Among widely-known PUF variants, the Arbiter PUF based on digital building blocks has a regular structure and low hardware footprint; however, it is susceptible to machine learning-based modeling attacks. In this work, we propose a new low-power and reliable PUF, based on the hybrid current mirror inverter (CMI). The proposed PUF circuit exploits the process variation-induced randomness in the CMI circuits to generate instance-specific challenge-response characteristics. The current mirror ensures the stability of the overdrive voltages across all the transistors in the CMI blocks, thereby improving the reliability of the proposed PUF. The PUF circuit was simulated at a 45 nm CMOS technology node, and its major performance metrics were evaluated. The simulation results demonstrated that the proposed PUF has excellent performance metrics while having low hardware and resource footprint. In addition, the proposed PUF demonstrated robustness against machine learning-based model-building attacks.

The growing importance of the internet has intensified the risks to network systems, necessitating enhanced security measures. Intrusion Detection Systems (IDS) play a crucial role in protecting against hostile activity, ensuring the integrity of data, and maintaining business continuity. By utilizing machine learning, contemporary intrusion detection systems (IDS) provide instantaneous monitoring and efficient avoidance of potential threats. This article presents an IDS that utilizes a machine learning framework to specifically identify attacks in the UNSW-NB15 and NSL-KDD datasets. Data preprocessing involves handling missing values and applying standard scaler normalization to achieve uniform feature scaling. The Binary Bat Algorithm (BBA) improves model efficacy through feature extraction. By creating artificial samples, SMOTE-ENN addresses class imbalances and enhances the model’s ability to make accurate predictions across different classes. The attacks are classified using a Random Forest (RF) model, which achieves an accuracy of 97.3% while reducing the time required for training. The results highlight the exceptional performance of our IDS in comparison to conventional approaches, providing a potential option for comprehensive intrusion detection in network systems.

Stock market forecasting is one of the most exciting areas of time series forecasting both for the industry and academia. Stock market is a complex, non-linear and non-stationary system with many governing factors and noise. Some of these factors can be quantified and modeled, whereas some factors possess a random walk behavior making the process of forecasting challenging. Various statistical methods, machine learning, and deep learning techniques are prevalent in stock market forecasting. Recently, there has been a paradigm shift towards hybrid models, showing some promising results. In this paper, we propose a technique that combines a recently proposed deep learning architecture N-BEATS with wavelet transformation for improved forecasting of future prices of stock market indices. This work uses daily time series data from five stock market indices, namely NIFTY 50, Dow Jones Industrial Average (DJIA), Nikkei 225, BSE SENSEX, and Hang Seng Index (HSI), for the experimental studies to compare the proposed technique with some traditional deep learning techniques. The empirical findings suggest that the proposed architecture has high accuracy as compared to some traditional time series forecasting methods and can improve the forecasting of non-linear and non-stationary stock market time series.

The convergence of the Internet of Things (IoT) and Artificial Intelligence (AI) is significantly transforming the landscape of future networking. The Internet of Things (IoT) is a technological paradigm that encompasses embedded systems, wireless sensors, and automation, facilitating the integration of various applications ranging from smart homes to wearable devices. In addition, the advent of artificial intelligence (AI) amplifies this influence by providing data-driven analytics, optimising processes, and presenting novel opportunities for growth. Nevertheless, the widespread adoption of devices within Internet of Things (IoT) networks gives rise to apprehensions regarding increased energy consumption. In order to ensure the longevity of network operations, it is imperative to employ energy-efficient protocols for sensor nodes that possess limited power resources. One example of a protocol that demonstrates this concept is the Low Energy Adaptive Clustering Hierarchy (LEACH) protocol. This protocol effectively divides networks into clusters and dynamically adjusts the cluster heads to optimise the transmission of data to the base stations. Our study enhances the LEACH protocol by incorporating digital twin simulation, thereby enhancing the efficiency of IoT systems. Virtual network models and AI analytics are employed to assess energy consumption and performance. Cache nodes play a crucial role within this framework as they collect data from cluster heads in order to transmit it to the base station. By leveraging artificial intelligence (AI) and simulation techniques, we are able to improve the energy efficiency and reliability of the Internet of Things (IoT) systems. The findings indicate a significant reduction of 83% in non-functioning nodes and a notable increase of 1.66 times in energy levels of nodes compared to conventional approaches. This study highlights a potential direction for energy-efficient, AI-enhanced Internet of Things (IoT) networking through the utilisation of the Low-Energy Adaptive Clustering Hierarchy (LEACH) protocol.

With the rapid growth of the Internet of Things technologies, individuals’ security and privacy concerns increase. There has been a leaky ecosystem webbed by the presence of sensors and devices around us. It has become a threat for the users and raises concerns for its widespread applicability. Many works have been proposed in a similar direction, building innovations for enhancing security. Meanwhile, Physical unclonable functions (PUFs) and Random Number Generators (RNGs) have proved to be the most useful for building security applications, especially in resource restraining devices. The security protocols, including identification, authentication, and key agreement, can be developed using PUFs, while RNGs can produce ephemeral keys and nonces. The true random number generator (TRNG) and PUF as a re-configurable circuit reduce the hardware cost and become favorable in the resource-constrained environment. This paper proposes a low-cost re-configurable TRNG–PUF, which can be used as both TRNG and PUF to harness their benefits. The proposed re-configurable TRNG–PUF is implemented on FPGA to verify the design. We evaluate the proposed scheme against various parameters and compare them with the existing designs, which proves our claim of better performance.

The rising demand for electricity and its essential nature in today's world call for intelligent home energy management systems that can reduce energy usage. This article aims a novel way to develop a learning system that can learn from experience to shift loads from one time instance to another and achieve the goal of minimizing the aggregate peak load. Specifically, this article proposes a deep reinforcement learning model for demand response where the virtual agent learns the task like humans learns a task. The agent gets feedback for every action it takes in the environment; these feedbacks will drive the agent to learn about the environment and take much smarter steps later in its learning stages. The proposed approach outperformed the state of the art mixed integer linear programming for load peak reduction. Other key contribution is the design of an agent to minimize both consumer electricity bills and system peak load demand simultaneously. The proposed model was analyzed with loads from five different residential consumers; the method increases the monthly savings of each consumer by reducing their electricity bill drastically along with minimizing the peak load of the grid.

The spread of surveillance cameras has necessitated the monitoring of large quantities of surveillance video feeds. A manual monitoring system is near impossible due to the large man-hour requirements. Recently, automatic abnormal activity detection has been an area of interest among researchers. A spatio-temporal feature, histogram of optical flow orientation and magnitude (HOFM), has produced impressive ability in detecting abnormal activities. The authors propose a novel non-uniform spatio-temporal region resembling parallelepipeds, from which they extract the HOFM features. Autoencoders can be configured to detect abnormal patterns. The authors have used these abilities of the autoencoders to detect abnormalities in the HOFM features extracted from their novel spatio-temporal regions of the video feeds. The autoencoders are trained on the HOFM features of the videos containing no abnormalities. The autoencoders are then fed with the HOFM features of the videos to be tested for abnormal activities, and these are detected based on the abilities of the autoencoders to reconstruct these features. The proposed method is tested on the standard abnormality detection datasets: UCSD Ped1, UCSD Ped2, Subway Entrance, Subway Exit, and UMN.

A single-view face detector and a novel face recognition method based on the aggregate channel feature (ACF) that work at real-time speeds, suitable in a computing resource-constrained setting are presented in this work. The four stage tree-based face detector is trained on a subset of the AFLW dataset. The face detection performance is analysed using the AFW dataset. The face recogniser uses ACF features along with classification algorithms, either SVM or MLP. The face recogniser is trained and tested on the GATech Face dataset. Our face detector displays comparable performance against the state of the art while working at 29.8 fps. The face recogniser achieves a level of performance that is competitive with other state of the art works. The effect of PCA-based dimension reduction of ACF features on face recognition performance is also studied in this work.

With the advent of readily and widely available applications based on deepfake technology, several cybersecurity threats are on the rise. It is challenging to curtail these threats as deepfakes are realistic and very difficult to detect. The present work proposes a Fake Detection Tool (FDT) that streamlines the procedure of fake detection by incorporating various manipulation techniques and aids users in detecting and visualizing the same. The tool is also integrated with Twitter for streaming facial image posts based on hashtags. It provides an output dataframe and presents statistics of virality, sentiments, etc, using pie charts for better visualization. The proposed tool uses a wide variety of large-scale datasets for training to deal with the fakes in the wild and deploys models that are at par with the cutting-edge models. It is an efficient, user-friendly, and freely available software for fake detection. The source code of the FDT package toolkit is available at https://github.com/surbhiraj786/GUI_Fake-Detection.

Recently cognitive radio has been proposed to solve the problem of spectrum scarcity with the help of an opportunistic utilisation of unused spaces in the spectrum. The implementation of cognitive radio for real-time application faces with the challenge of real-time media access control (MAC) design. To this end the authors propose an energy efficient cognitive radio MAC protocol with two level of quality of service (QoS) support: two level quality of service provisioning in cognitive radio (TQCR)-MAC. The data traffic in the cognitive radio network is classified as real-time traffic and non-real-time traffic, with former having higher priority over latter. The proposed protocol utilises variable interframe spaces for implementing different priorities to data and also exploits the combination of multiple channels and time division multiple access (TDMA) scheme to further improve the QoS provisioning. The TDMA slots in different channels are used as the data carrier segment, in which a real-time data can take over the time slots reserved by non-real-time data. Thus, non-real traffic may suffer from starvation, and hence a guarantee on the non-real-time data is provided with the help of a starvation prevention algorithm. The performance analysis of the proposed MAC protocol is done against two similar MAC protocols, which shows that the performance is improved significantly, especially when the network is either biased towards real-time or non-real-time data traffic.

Spectrum sensing is an important task for the existence of cognitive radio. Accuracy of sensing is essential to eliminate interference to primary users. In a practical scenario, primary users may change its position and power frequently. A new scenario that may challenge cooperative spectrum sensing is considered here. An antenna selection scheme with multiple regions encoding for single node sensing and an adaptive weighted combining fusion rule for decision fusion are proposed to manage this scenario. The performance of the rule was analyzed in the new scenario under Rayleigh fading and compared with other fusion rules in the literature namely soft optimized linear combining (SOLC), soft equal weight linear combining, AND, OR, and voting. Probability of detection of the proposed approach was found better than SOLC (best among the rest) at low average SNR. Simulations showed that the probability of detection of proposed approach was nearly 10%better than SOLC at probability of false alarm Pf=0.01 and gradually became equal at Pf=0.15, and it maintains it as Pfgoes up. Computational complexity of these approaches were measured as the time consumed in making a decision, and it was found that the proposed approach consumed only 20%of the time than that of SOLC approach. Effect of number of neighbors on the performance of these approaches were also analyzed. Adaptive nature of this rule makes it suitable to perform well in any scenario under distributed sensing. This article proposes an antenna selection scheme with multiple regions encoding for single‐node sensing and an adaptive‐weighted combining fusion rule for decision fusion in cognitive radio networks. This method is able to make right spectrum decisions in a scenario where the cognitive radio terminals are located around the boundary of a primary user. Higher probability of detection and lower computational time make this method superior to other optimised approaches used for decision fusion.