Typically, load forecasting models are trained in an offline setting and then used to generate predictions in an online setting. However, this approach, known as batch learning, is limited in its ability to integrate new load information that becomes available in real-time. On the other hand, online learning methods enable load forecasting models to adapt efficiently to new incoming data. Electricity Load and Price Forecasting (ELPF) is critical to maintaining energy grid stability in smart grids. Existing forecasting methods cannot handle the available large amount of data, which are limited by different issues like non-linearity, un-adjusted high variance and high dimensions. A compact and improved algorithm is needed to synchronize with the diverse procedure in ELPF. Our model ELPF framework comprises high/low consumer data separation, handling missing and unstandardized data and preprocessing method, which includes selecting relevant features and removing redundant features. Finally, it implements the ELPF using an improved method Residual Network (ResNet-152) and the machine-improved Support Vector Machine (SVM) based forecasting engine to forecast the ELP accurately. We proposed two main distinct mechanisms, regularization, base learner selection and hyperparameter tuning, to improve the performance of the existing version of ResNet-152 and SVM. Furthermore, it reduces the time complexity and the overfitting model issue to handle more complex consumer data. Furthermore, numerous structures of ResNet-152 and SVM are also explored to improve the regularization function, base learners and compatible selection of the parameter values with respect to fitting capabilities for the final forecasting. Simulated results from the real-world load and price data confirm that the proposed method outperforms 8% of the existing schemes in performance measures and can also be used in industry-based applications.

Recent years has seen a tremendous increase in processing requirements of present-day embedded system applications. Embedded systems consist of multiple processing elements (PEs) connected to each other using different types of interfaces. Many complicated tasks are accomplished by embedded systems in varied settings, which may introduce errors during inter-processor communication. Testing such systems is tremendously difficult and challenging from testing non-real time systems. A major part of testing real time embedded systems involves ensuring accuracy and timing in synchronous inter-process communication More specifically, the synchronization and inter-processor communication of real-time applications makes testing a challenging task and due to the demand for higher data rate increases, day-by-day, making testing of such systems even more complex. This paper presents a novel frame work that uses multiple instances of simulators with physical high-speed serial interfaces to emulate any real time embedded system communication. The framework presents a testing technique that detects all faults related to synchronization of high-speed synchronous serial interfaces in a systematic manner. The novelty of our approach is to simulate communication across multiple processors in a simulation environment for detecting and localizing bugs. We verify this framework using a case study consisting of an embedded software defined radio (SDR) system. The test results show the applicability of our approach in fixing bugs that relates to synchronization issues that otherwise are very hard to find and fix in very complicated systems, such as SDR.

Embedded systems are commonly used to meet the rapid growth in processing requirements of today's embedded system applications that range from simple home appliances to complex big architecture systems. An embedded system consisting of multiple processing elements (PEs) linked together using various types of interfaces. Many complicated tasks can be accomplished by them in different working environments. Synchronization is the typical problem with which embedded software are confronted. The processing elements require synchronization to preserve system integrity while performing the applications. Testing multi PEs-based embedded system is a challenging task; where most of the intricacies lie in testing realtime interprocessor communication i.e. synchronous or asynchronous in term of accuracy and timing. This paper is intended to provide an outline of issues affected by the need for considering synchronization on multi PEs embedded system. The paper proposes an approach that simulates possible synchronization issues across multi PEs in real-time communication. The results show that the simulation of synchronization issues on a multi PEs embedded system is a valuable approach towards creating a testing framework for embedded systems.

The high rate of increase in the deep learning tasks as well as heterogeneous computing systems necessitates compilers that achieve low compile time and high performance. The current state of the art in the use of tensor compilers is based on exhaustive search, a slow and prohibitive process, or heuristics at the expense of generality and quality of optimisation. The paper presents the Tensor Language Model (TLM), a generative framework of a compiler that redefines the optimisation of tensor programmes as a language modelling problem. TLM is also based on a GPT-2 architecture that is pre-trained on millions of tensor programs coded as compact tensor code capturing operator graphs, hardware metadata, and reconfiguration choices as a sequence of tokens. The model creates optimised schedules of tensors, thus avoiding any search or reinforcement learning at run time. Experimental results using ResNet-50, BERT, GPT-2 and LLAMA-7B indicate that TLM compiles up to 61 times faster than search-based compilers (e.g., Ansor, MetaSchedule) and is up to 2.25 times faster than heuristic models (e.g. Roller), and has similar runtime efficiency. TLM then selects a trade-off seemingly traditional between compilation time and execution performance, a scalable, hardware-agnostic and reproducible generative paradigm of next-generation deep learning compilers.

Recent advances in machine learning and deep learning have greatly improved how we detect plant diseases, making diagnoses more accurate, faster, and easier to scale. However, many existing solutions depend on large, pretrained models that need powerful hardware, which limits their use in the field, especially in areas with limited resources. To tackle this, we designed a custom lightweight convolutional neural network (CNN) built from scratch using 20,000 carefully selected images from the PlantVillage tomato dataset. Our model uses Squeeze-and-Excitation (SE) blocks and Swish activation functions to boost performance, reaching an accuracy of 97.7% while using far fewer computing resources than many standard models. We applied strong image preprocessing steps like resizing, normalization, and various augmentations, along with regularization methods such as dropout and weight decay to reduce overfitting. Compared to popular models, our network performs better than object detection model (91.9% accuracy) and is nearly on par with the transformer-based model (98.0% accuracy), but is far more practical for mobile devices and edge applications. In the future, we plan to add features for grading the severity of diseases, create an easy-to-use interface, and expand our model to cover more crops.

Cloud computing has emerged as a powerful new technology. The processing and computation power embedded in the cloud technology is not only flexible but also infinitely scalable and cost effective. Service oriented architecture (SOA) is a perfect stage for cloud computing. SOA has allowed customers and organizations to achieve cloud computing and reap its benefits that would not have been possible through any other architecture. This paper discusses the concept and importance of service oriented cloud computing by highlighting possible architectures, their benefits and critical success factors.