Shift-invariant neural network for image processing: learning and generalization

The rise of deep learning has accelerated the application of neural network in image processing. As the nearest neural network model to human biological cell system, spiking neural network (SNN) has shown great potential in image processing and pattern recognition. In this paper, a method of applying spiking neural network to image processing is proposed. Recent advances in neuroanatomy have provided favorable conditions for the application of local connections. By analyzing the connection modes of different sparsity of spiking neurons in space, spiking neural network can effectively perform robust image processing tasks. The experiment shows that the pre-set spiking neural network can acquire image features quickly and effectively without training.

Neural networks usually work adequately on small problems but can run into trouble when they are scaled up to problems involving large amounts of input data. Circuit Complexity and Neural Networks addresses the important question of how well neural networks scale - that is, how fast the computation time and number of neurons grow as the problem size increases. It surveys recent research in circuit complexity (a robust branch of theoretical computer science) and applies this work to a theoretical understanding of the problem of scalability. Most research in neural networks focuses on learning, yet it is important to understand the physical limitations of the network before the resources needed to solve a certain problem can be calculated. One of the aims of this book is to compare the complexity of neural networks and the complexity of conventional computers, looking at the computational ability and resources (neurons and time) that are a necessary part of the foundations of neural network learning. Circuit Complexity and Neural Networks contains a significant amount of background material on conventional complexity theory that will enable neural network scientists to learn about how complexity theory applies to their discipline, and allow complexity theorists to see how their discipline applies to neural networks.

Most of the proposed neural networks for fault diagnosis of systems are multilayer perceptrons (MLP) employing the popular backpropagation (BP) learning rule. It has been shown that the backpropagation algorithm usually takes a long time for convergence and sometimes gets trapped into local minimum. The algorithm requires the architecture to be fixed initially (i.e. the number of hidden units) before learning begins. Final network size is obtained by repeated trials. When the size of the training set is large, especially in the case of fault diagnosis, such a repeated training consumes a large amount of time and sometimes it can be frustrating. Thus there is a need of a good neural network architecture that decides its size automatically while learning the input/output relationships and must posses reasonably good generalization. Neural networks based on radial basis functions (RBF) have emerged as potential alternatives to MLPs. RBFs have a simple architecture and they can learn the input/output relations fast compared to MLPs. In this paper we present a constructive neural network based on radial basis functions (RBF) due to Fritzke for classification of fault patterns in a model power system. The performance of this neural network with traditional BP network and nonconstructive RBF network in terms of size, learning speed and generalization are presented.

In today’s world, the need for computers and their logics is critical for the development of any system. This paper is for those people who have little or no knowledge about Artificial Neural Networks (ANNs). Various advances have been made in creating intelligent systems, some enlivened by biological neural networks. Analysts from numerous logical orders are outlining artificial neural networks (ANNs) to take care of a variety of problems in design recognition, prediction, optimization, associative memory, and control. And will be discussing some of the algorithms of ANNs such as image segmentation using edge detection, image enhancement using multiscale retinex and technique combining sharpening and noise reduction and some practical applications such as diagnosing liver disease and face detection systems.

Multiple disorder diagnosis with adaptive competitive neural networks

To deepen the understanding of the human brain, many researchers have created a new way of analyzing neural data. In many previous studies, researchers have examined neural networks from a macroscopic point of view, based on neuronal firing patterns. On the contrary, we have studied neural networks locally, in order to understand their communication strategies. To understand information processing in the brain, we simulated the firing activities of neural networks in a 9 × 9 two-dimensional neural network to analyze spike behavior. In this research study, we used two kinds of learning processes. As the main learning process, we implemented the learning process to identify the stimulated position. As the subsidiary one, we implemented Hebb’s learning rule which changes weight between neurons. Three channels with transmission and reception were preset, each of which has a different distance and direction. When all three channels succeeded in identifying the source stimulation in the receiving neuron group, it was regarded as an overall success and the learning was termed as successful. Furthermore, in order to see the effect of the second learning procedure, we elucidated the average of necessary learning times in each channel type and compared the firing propagation time of the first trial and an overall successful trial, in each channel. We found that the firing path after learning is shorter than the firing path before learning. Therefore, we deduced that Hebb’s rule contributes to shortening the firing path. Thus, Hebb’s rule contributes to speeding up communication in a neuronal network.

The Sixth International Symposium on Neural Networks (ISNN 2009)

This paper delves into the realm of Feedforward Neural Networks (FFN) and explores their application in image processing.The research begins with an introduction to FFNs, elucidating their fundamental concepts and underpinning principles.The FFN's architecture, comprising input, hidden, and output layers, along with interconnected nodes and weighted connections, is detailed to provide a comprehensive understanding of its structural framework.Moreover, the paper investigates the usage of FFNs in image processing, highlighting their role in various tasks such as image classification, object detection, image recognition, and image generation.Through a critical analysis, this research explores the advantages of employing FFNs in image processing, emphasizing their capability to learn complex patterns, adaptability to large datasets, and ability to generalize well to new data.Conversely, the paper also examines the limitations and disadvantages associated with FFNs in image processing, including issues related to overfitting, computational complexity, and dependency on extensive data for effective training.Furthermore, the study focuses on the implementation of FFNs in image processing applications.It discusses specific methodologies and techniques utilized to integrate FFNs into image-based tasks, showcasing their effectiveness in enhancing image quality, extracting features, and performing intricate image-based computations.Additionally, this paper presents case studies and realworld examples where FFNs have been successfully deployed in image processing applications.These instances illustrate the practical implications and the significance of FFNs in domains such as medical imaging, satellite image analysis, facial recognition, and autonomous vehicles.In conclusion, the research sheds light on the potential and challenges of using Feedforward Neural Networks for image processing tasks.By comprehensively analyzing their architecture, usage, advantages, and drawbacks, this paper aims to contribute to the deeper understanding and informed utilization of FFNs in the field of image processing.

For the path planning problem of autonomous underwater vehicles (AUVs) in 3-dimensional (3-D) estuary environments, traditional methods may encounter problems due to their high computational complexity. In this paper, we proposed a dynamic neural network to solve the AUV path planning problem. In the neural network, neurons get input from the environment, locally interact with the neighbors and update neural activities in real time. The AUV path is then generated according to the neural activity landscapes. Stability, computational complexity of the neural network, and optimality of the generated path are analyzed. AUV path planning in 3-D complex environments without currents, with constant currents, and with variable currents are studied through simulations, which demonstrate the effectiveness of this approach.

In this paper we propose a new method for the detection and evaluation of exudates in retinal images. In the learning phase we focus on selecting efficient features that can uniquely identify the exudates. In this process we develop a neural network for image processing. We then further extend and train the neural network to detect and evaluate the exudates. The efficiency of the proposed method is compared with other similar researches, improvements in accuracy being observed.

Comparative analysis of convolutional neural network and recurrent neural networks for image processing and data analysis

Network motif identification is significant in that motifs generally reflect functionalities. However, the task is greatly challenging due to diverse patterns of motifs possibly existent in a network and high computation complexity for large scale networks. In this study, we propose a network motif identification method, FCMD, based on clustering subgraphs according to their local structures. By modeling local-topological feature of a network with a feature vector, the approach maps motif identification problem into a clustering problem of feature vectors in a feature space, which greatly reduces computation complexity hence facilitates very large scale networks. Experiments on 8 real networks, including biochemical network, neural network, electronics circuit network, et al., indicates that the proposed method is very effective in motif detection with computation complexity approximately independent to the scale of the network.

While Network Architecture Search (NAS) and network pruning have recently re-emerged as powerful techniques to design more effective networks with less parameters, their focus has been on improving deep neural network models with millions of parameters. However, the number of connections and the connectivity pattern between layers also influences the learning performance of shallow two-layer networks, such as the classic Restricted Boltzmann Machine (RBM). This work presents a comprehensive study of the connectivity space on the learning performance of RBMs using the bars and stripes (BAS) model and the MNIST dataset. For BAS, the main findings show that pattern and number of connections play a fundamental role on the learning performance, while for MNIST the number of connections is more important for the considered connection structures. Moreover, the fully connected network is outperformed by different patterns in both scenarios, indicating the importance of also designing more effective connectivity patterns for simple two-layer neural networks.

Based on the theoretical basis of quantum computation, entanglement has been further explored as one of the key resources. An outstanding problem in quantum computation is the detecting of entanglement, but no closed-form algorithm exists. According to an advanced matrix rearrangement approach for detecting multipartite quantum states, we give some examples to verify the effectiveness and practicability of this new criterion. Meanwhile, this paper shows how techniques from the quantum neural network can be utilized to detect entanglement. Two models, known as the discrete-variable and the continuous-variable quantum neural network, are applied to solve the separability problem. We demonstrate the utility that such a discrete-variable quantum neural network can be trained to detect the entanglement with great exactness. And the complexity of network and computation is greatly reduced. Besides, the continuous-variable quantum neural network is applied for detecting two-qumode Gaussian state. Compared with the results of the traditional neural network, this deep quantum neural network demonstrates its capability and adaptability as well.

Identification of Hydroponics in an Advanced Control System of the Greenhouse