Dataflow accelerators feature simplicity, programmability, and energy-efficiency and are visualized as a promising architecture for accelerating perfectly nested loops that dominate several important applications, inc...
详细信息
Dataflow accelerators feature simplicity, programmability, and energy-efficiency and are visualized as a promising architecture for accelerating perfectly nested loops that dominate several important applications, including image and media processing and deep learning. Although numerous accelerator designs are being proposed, how to discover the most efficient way to execute the perfectly nested loop of an application onto computational and memory resources of a given dataflow accelerator (execution method) remains an essential and yet unsolved challenge. In this paper, we propose dMazeRunner - to efficiently and accurately explore the vast space of the different ways to spatiotemporally execute a perfectly nested loop on dataflow accelerators (execution methods). The novelty of dMazeRunner framework is in: i) a holistic representation of the loop nests, that can succinctly capture the various execution methods, ii) accurate energy and performance models that explicitly capture the computation and communication patterns, data movement, and data buffering of the different execution methods, and iii) drastic pruning of the vast search space by discarding invalid solutions and the solutions that lead to the same cost. Our experiments on various convolution layers (perfectly nested loops) of popular deep learning applications demonstrate that the solutions discovered by dMazeRunner are on average 9.16x better in Energy-Delay-Product (EDP) and 5.83x better in execution time, as compared to prior approaches. With additional pruning heuristics, dMazeRunner reduces the search time from days to seconds with a mere 2.56% increase in EDP, as compared to the optimal solution.
Chillies are a high-value export commodity. One of the chilli products that has a high export value, such as to the European market, is dried red chillies. Dried chillies' quality is the main parameter that must b...
Chillies are a high-value export commodity. One of the chilli products that has a high export value, such as to the European market, is dried red chillies. Dried chillies' quality is the main parameter that must be maintained if this commodity is to be continuously exported. This study aimed to precisely and accurately classify the quality of dried chillies using physical parameters of the length and color of chillies based on digital imageprocessing. In this research, the quality of dried red chillies was classified using a combination of digital imageprocessing and artificialneuralnetworks (ANN). The quality parameters of dried chillies used as inputs were chilli length, mean energy, mean a*, mean blue, and mean contrast. This study used 150 dried chillies for training data and 36 samples for test data. The classification of the quality of dried red chilli was divided into three classes, which were extra class, class I, and class ii. The result of this study was an artificialneural network structure consisting of 5 input layer cells, 16 hidden layer cells, and two output layer cells. The testing of the system using 36 testing data that determined the values of dried chillies reached 94.4%.
Analysis of data and synthesis for hyper spectral imaging (HSI) is a new branch of remotely sensed data and planet surveillance technologies. Classification techniques with the help of deep learning for Land cover hav...
Analysis of data and synthesis for hyper spectral imaging (HSI) is a new branch of remotely sensed data and planet surveillance technologies. Classification techniques with the help of deep learning for Land cover have recently been a popular research topic, so these techniques are employed in a variety of applications, including agricultural, environmental analysis, military surveillance, mineral extraction and urban exploration. Despite the fact that Convolutional neuralnetworks (CNNs) have attracted a lot of attention and produced impressive results in a range of scene classification tasks, their high computational and storage costs restrict them from being employed in real-time remote sensing applications. Traditional CNN-based techniques, on the other hand, focus on producing scene representation by processing features of original image or from CNN, neglecting the fact that texture pictures or each layer of CNNs convey information that is distinct. To efficiently investigate the characteristics and avoid the above-mentioned drawbacks, a pretrained DNN model-based method is offered. The performance of two different hyperspectral classifiers for land use/land cover classification is compared in this study. The HSI images were classified using Multilayer Perceptron artificialneuralnetworks and Support Vector Machines. The experimental results, which were carried out on the Live Aerial image Hyperspectral Datasets, revealed that classification accuracy is analyzed with and without feature reduction and it is found that with feature reduction performance of proposed algorithm is improved with significant margin with respective to existing aerial image dataset.
Computer Vision and Natural Language processing methods have been employed in artificial Intelligence to describe the content of a picture. It is useful in assisting the visually impaired, as well as in self-driving a...
Computer Vision and Natural Language processing methods have been employed in artificial Intelligence to describe the content of a picture. It is useful in assisting the visually impaired, as well as in self-driving automobiles and other applications. It is not simpleto describe an image, in order to do so, we must use appropriate phrase structure. However, by utilizing deep learning techniques, the process becamesimple and straightforward. CNN (pre-trained Xception model) and LSTM were the algorithms we employed. Convolutional neuralnetworks are deep learning algorithms thatextract features from an input image. The extracted features are sent to the next layer,which is then fed to LSTM. In LSTM, we use a sequence processor to get sentences in order, and then we construct the final caption of the input image through LSTM. It is a type of deep convolutional neural network with depth-wise separable convolutions. We used Flickr_8k *** used Google Colab for training and testing the model. An image will be given as input, the caption will be generated describing the image and the caption is read out(converted to audio) when selected by the user.
The categorization of medical images, which includes picture segmentation and detection, appears to be a key task in the field of artificial intelligence. Support vector machines (SVM), convolutional neuralnetworks (...
The categorization of medical images, which includes picture segmentation and detection, appears to be a key task in the field of artificial intelligence. Support vector machines (SVM), convolutional neuralnetworks (CNN)have all been applied to the classification of images. Given the capabilities and adaptability of deep learning across a range of machine learning applications, CNN has emerged as one of these methods that has been particularly successful in recent years. This research offers a thorough analysis of CNN’s fundamental ideas and we apply to medical imageprocessing. the core ideas of CNN are outlined, emphasising its capacity to recognise spatial relationships in images and learn hierarchical characteristics. We examine the architecture of CNN and its essential elements, comprising fully connected, pooling, and convolution layers. Further investigation is done into how deep learning can be used to classify medical images more accurately. We offer a variety of CNN models, from basic ones like LeNet-5 to more complex ones like AlexNet, VGGNet, Google Net, and ResNet. Each model’s unique qualities and innovations are discussed, offering insight on how each model contributes to medical picture categorization problems. The application of imageprocessing methods for task augmentation and picture refining is covered in depth in this work. It is addressed how methods like image enhancement, denoising, and feature extraction could enhance the precision and calibre of medical images. Researchers can learn more about the developments and potential of imageprocessing algorithms and CNN-based approaches in the field of medical imaging. This research aims to increase classification accuracy and help the development of reliable and effective medical imaging systems by advancing our understanding of the use of CNN for imageprocessing in relation to applications in medicine.
This research delves into the combination of SHapley Additive exPlanations (SHAP) with Convolutional neuralnetworks (CNNs), highlighting the role of SHAP value in enhancing the transparency and interpretability of AI...
This research delves into the combination of SHapley Additive exPlanations (SHAP) with Convolutional neuralnetworks (CNNs), highlighting the role of SHAP value in enhancing the transparency and interpretability of AI models. CNNs, known for their efficacy in imageprocessing and pattern recognition, often suffer from a lack of transparency due to their complex layered structures. This ’black box’ nature poses significant challenges in understanding and trusting artificial Intelligence (AI) driven decisions and AI ethics, especially in critical applications such as healthcare and autonomous systems, emphasizing the necessity for interpretable AI systems. Grounded in cooperative game theory, SHAP offers a novel approach to deciphering these complexities by quantifying the contribution of each input feature to the model’s output. Therefore, this research first elucidates the principles of CNNs, by discussing different types of layers applied in CNN and the challenge of interpreting CNN results from CNN’s nature. Then the mechanisms of SHAP and the functionality of SHAP values are explored. Eventually, the application of SHAP on advanced AI transparency and accountability is illustrated based on the iris dataset, which proved to be helpful when analyzing the model after quantifying the influence of input features.
Currently, in many applications, humans are often used to translate satellite images into useful data. Based on the experience level of the humans, the accuracy of the obtained data is varied. This research presents a...
Currently, in many applications, humans are often used to translate satellite images into useful data. Based on the experience level of the humans, the accuracy of the obtained data is varied. This research presents a method that adopts one of the artificial intelligence methods, specifically neuralnetworks, to produce useful data that can be used in decision-making to detect water hyacinth weed. The proposed approach will enable scientists to rapidly recognize the world’s worst aquatic weed, Water Hyacinth. This weed currently creates serious agricultural and navigation problems in Iraq. It affects irrigation, water flow, water use and navigation. Consequently, the proposed approach can enhance in maximizing the economic factor, minimizing CO2 emissions and also minimize land degradation.
Face recognition is one of the most active areas of research from the past two decades. Attempts are being made to understand how a human recognizes another human face. It is widely accepted that facial recognition ca...
Face recognition is one of the most active areas of research from the past two decades. Attempts are being made to understand how a human recognizes another human face. It is widely accepted that facial recognition can be based on structural information and nonstructural / spatial details. In the present study, he is applying differential observations using Eigen / docking characteristics of many built-in facial features and artificialneuralnetworks. The proposed method aims to obtain a facial feature by reducing facial features such as eyes, nose, mouth, and face depending on the importance of facial features. The face recognition system developed in this paper will inform the human face and assess the current percentage of accuracy. Therefore, this work is for human facial recognition and includes a percentage of facial expressions. The implementation of this function also offers many applications such as photography, bio-metric in bank Lockers, etc.
暂无评论