Real-time Electrocardiogram (ECG) classification on wearable devices is critical for cardiac monitoring but limited by power and bandwidth constraints. We present the first deployable ECG arrhythmia classification on commercially available neuromorphic Akida AKD1000 hardware, converting a 2D convolutional neural network (CNN) into a 4-bit spiking neural network (SNN). This enables the transmission of only 3-bit semantic alerts over the air, with a decoding rate exceeding 99% in a simulated Rayleigh fading channel. Evaluated on the MITBIH Arrhythmia, our quantization-aware-trained model attains 99.02% accuracy across five heartbeat classes while consuming 2.04 μJ per inference and 1.66 ms latency at 100 MHz. Compared to streaming raw ECG, the proposed system reduces wireless payload by >99%. These results underline the feasibility of battery-powered, always-on cardiac monitoring for 5G/6G telehealth and the potential of neuromorphic computing to enable energy-efficient, accurate, real-time ECG analysis directly on edge devices, paving the way for advanced wearable health monitoring systems and applications in remote healthcare leveraging semantic communication principles.

Edge AI has become the cornerstone of many applications. Yet, progress is limited by the large complexity of training a deep neural network (a DNN). hyperdimensional computing (HDC) is positioned as an alternative approach for Edge AI that is compact enough to enable training. The main challenge for an HDC model is to maintain its key features while balancing high inference accuracy with efficiency. A simple binary HDC model lacks accuracy, while the computational complexity of a floating-point model is too high. This work presents FixedHD, a novel 16-bit fixed-point HDC model enabling training at the Edge. FixedHD achieves an accuracy similar to floating-point model while lowering computational complexity. The model is supported by a customized RISC-V processor tailored to speedup both training and inference. The processor is extended with advanced HDC-specific instructions, a vector unit to utilize HDC’s parallel nature, and, for the first time, approximate computing to exploit its robustness. Further, memory requirements are reduced by quantizing mathematical functions and reducing the large HDC encoding matrix by up to 390 x. Compared to the baseline processor, inference and training are accelerated on average by 6.9 x and 3 x, respectively. The energy consumption is reduced by 4.6 x and 1.9 x at the cost of an increase in area by 45 %. The inference accuracy remains at the high level of floating-point models despite the heavy quantization and approximation.

Integration between Quality Control and Industry 4.0 was explored seeking to obtain the best quality of the final product. This project aims to create an algorithm that creates a large data set that will train the Machine Learning model to achieve Automated Quality Control in Industry 4.0. The Methodology applied started with Data Acquisition by capturing a video of the fruits on the conveyor, afterwards applying Video Processing by cutting this video into frames to obtain raw data, then applying Image Processing on the raw data to clean it using Color Detection and Contour and Size Detection approaches, these methods resulted in the creation of two datasets of fresh and rotten oranges, with a total of 1700+ images of oranges. This paper presents the collection of a large dataset in addition to cleaning this data and make it ready for training and testing, Also to create an algorithm that creates large datasets of crops automatically

Large Intelligent Surfaces are essential components in upcoming 6G networks due to their impressive capabilities and energy efficiency. To enhance network coverage, we introduce relays that can overcome distance and obstacle limitations. The focus is on integrating Large Intelligent Surfaces with amplify-and-forward relay techniques to facilitate communication between a base station and a full-duplex user, even in the presence of co-channel interference from device-to-device pairs nearby. To optimize data transmission, closed-form expression is derived that calculates the optimal power required to achieve maximum data rates while considering challenges like self-interference and co-channel interference from device-to-device communications. Also, quadratic transform for the non-convex problem is derived analytically to be able to solve the problem using CVX as well. Additionally, a neural network approach is proposed that accurately predicts the minimum power needed to achieve the maximum data rates. This neural network model proves to be a promising alternative with much less complexity.

Large Intelligent Surfaces are key technologies for the upcoming 6G networks for their high capabilities and low power consumption. To further extend the coverage, relays can be employed to overcome distances and obstacle limitations. In this paper, we integrate the large intelligent surfaces technology with the decode-and-forward relay to aid the communication of a base station with a full-duplex user in the presence of device-to-device communicating devices. A closed form expression for the optimum power to achieve maximum data rate is derived in the presence of self-interference and co-channel interference from the device-to-device communications. A neural network approach is proposed, that accurately predicts the minimum power required to achieve maximum data rates. Results show that the neural network is a promising alternative to avoid complex calculations.

Abstract Network intrusion detection systems (NIDS) are the most common tool used to detect malicious attacks on a network. They help prevent the ever-increasing different attacks and provide better security for the network. NIDS are classified into signature-based and anomaly-based detection. The most common type of NIDS is the anomaly-based NIDS which is based on machine learning models and is able to detect attacks with high accuracy. However, in recent years, NIDS has achieved even better results in detecting already known and novel attacks with the adoption of deep learning models. Benchmark datasets in intrusion detection try to simulate real-network traffic by including more normal traffic samples than the attack samples. This causes the training data to be imbalanced and causes difficulties in detecting certain types of attacks for the NIDS. In this paper, a data resampling technique is proposed based on Adaptive Synthetic (ADASYN) and Tomek Links algorithms in combination with different deep learning models to mitigate the class imbalance problem. The proposed model is evaluated on the benchmark NSL-KDD dataset using accuracy, precision, recall and F-score metrics. The experimental results show that in binary classification, the proposed method improves the performance of the NIDS and outperforms state-of-the-art models with an achieved accuracy of 99.8%. In multi-class classification, the results were also improved, outperforming state-of-the-art models with an achieved accuracy of 99.98%.