FEEDFORWARD NEURAL NETWORK: A Review

Feedforward Neural Network (FFNN) is a surrogate of Artificial Neural Network (ANN) in which links amongst the units do not form a directed cycle. ANNs, akin to the vast network of neurons in the brain (human central nervous system) are usually presented as systems of interweaving connected "neurons" which exchange messages between each other. These connections have numeric hefts that can be adjusted and grounded on experience, enforcing adaptively on neural networks to inputs and learning capabilities. This paper presents a comprehensive review of FFNN with emphasis on implantation issues, which have been addressed by previous approaches. We also propose a theoretical model that exhibits potential superior performances in terms of convergence speed, efficient and effective computation and generality than state of the art models.

Journal of Mathematical Psychology, 1997

Journal of the Society of Dyers and Colourists, 1998

2020

In this paper, an overview of the artificial neural networks is presented. Their main and popular types such as the multilayer feedforward neural network (MLFFNN), the recurrent neural network (RNN), and the radial basis function (RBF) are investigated. Furthermore, the main advantages and disadvantages of each type are included as well as the training process.

I used to thought that BP Neural Networks algorithm belongs to supervised learning, however, after learned about Sparse Autoencoder algorithm, I realized it can also be used for unsupervised learning (use the unlabeled data itself as both input and output). BP neural networks is the base of a lot of other advanced neural networks algorithm, it is easy, but powerful. NEURON AND NEURAL NETWORKS Neural networks is built by certain amount of computational units, each of these unit is called a 'Neuron', the reason of using these biological words is that, neuroscientists told us how our brains work, and we are trying to mimic part of a brain, make machines more powerful. First we did, as usual, establish an ideal model which describe the neural system, and our definition of neuron is something like this: Neuron is an unit which have inputs and outputs, accompanies with each inputs, there is a weight that decide how this input value distributes, means what percentage of this input value will be taken by each neuron. For the above neuron, the input is W1 * X1 + W2 * X2 + W3 * X3 + 1 * b, sometimes, we also treat this bias b as W0, and set X0 always has the value +1, so the input can also be written as W T X. What about the output? The output is output = f(input), we call this f() as activation function, here, I'll use sigmoid function as the activation function (sure you can use tanh() as well, as I said in the Logistic Regression post), that is, O1 = O2 = O3 = sigmoid(W T X). If we have multi-neurons, we get a neural network, a simple example is like this: In this particular case, we say we have 3 layers in the network, 1 input layer (L1), 1 hidden layer (L2), and 1 output layer (L3). In both input layer and hidden layer, we have a " +1 " unit, we call these bias units, we can either count bias as one of the neuron in its layer or not counting it, this is depend on how you implement the neural networks algorithm. See the rightmost of this network, the output of output layer h W,b(x), means the value of output that calculated using current weight W and b, this is exactly the same as Linear regression and Logistic regression.

2005

The approach for discrete dynamical system modeling based on feed-forward neural network is considered. The proposed method is realized in two stages. The first one is training of the feed-forward neural network with given experimental data obtained for the designed system. The second stage is simulation with the neural network model. The variant of the method for feed-forward neural network training known as “backpropagation” is developed. The effective realization in matrix form of the proposed algorithm is given. This approach is successfully applied to the design problem in time domain for the recursive digital filter. Some results and graphical representations for lowpass recursive digital filters are given.

IEEE Transactions on Neural Networks, 2004

Lecturer 14-15-Artificial Neuron Networks 2 Artificial neural networks  Artificial neural network (ANN)  Inspired by biological neural systems, i.e., human brains  ANN is a network composed of a number of artificial neurons  Neuron  Has an input/output (I/O) characteristic  Implements a local computation  The output of a unit is determined by  Its I/O characteristic  Its interconnections to other units  Possibly external inputs 2/16/2023 2 Artificial neural networks  ANN can be seen as a parallel distributed information processing structure  ANN has the ability to learn, recall, and generalize from training data by assigning and adjusting the interconnection weights  The overall function is determined by  The network topology  The individual neuron characteristic  The learning/training strategy  The training data 3 Applications of ANNs  Image processing and computer vision  E.g., image matching, preprocessing, segmentation and analysis, computer vision, image compression, stereo vision, and processing and understanding of time-varying images  Signal processing  E.g., seismic signal analysis and morphology  Pattern recognition  E.g., feature extraction, radar signal classification and analysis, speech recognition and understanding, fingerprint identification, character recognition, face recognition, and handwriting analysis  Medicine  E.g., electrocardiographic signal analysis and understanding, diagnosis of various diseases, and medical image processing

An artificial neural network is a system based on the operation of biological neural networks, in other words, is an emulation of biological neural system.