China
Africa
Explore chapters and articles related to this topic
Use of UAVs in the Prevention, Control and Management of Pandemics
Published in Om Prakash Jena, Bharat Bhushan, Nitin Rakesh, Parma Nand Astya, Yousef Farhaoui, Machine Learning and Deep Learning in Efficacy Improvement of Healthcare Systems, 2022
Giuliana Bilotta, Vincenzo Barrile, Ernesto Bernardo, Antonino Fotia
In machine learning, particularly in deep learning, backpropagation is a widely used algorithm in training neural networks for supervised learning. Neural networks are characterized by the presence of loops that have a large impact on the learning and prediction capabilities of the network:The initialization of “weights” in a random manner;The propagation of the initial data with relative multiplication by the “weights”, and then passing the results through the activation function;The comparison of the obtained results with the supervised ones;The error assessment to understand the goodness of the “weights” adopted;The actual backpropagation phase of adjustment of the “weights” if necessary.
When the Physical Disorder of CMOS Meets Machine Learning
Published in Choi Jung Han, Iniewski Krzysztof, High-Speed and Lower Power Technologies, 2018
Xiaolin Xu, Shuo Li, Raghavan Kumar, Wayne Burleson
An important purpose of configuring the propagation delay of each element is to minimize the deviation of process variations of CMOS transistors. However, since all components in the CCATDC design are of nanometer magnitude, it is again impractical to measure such process variations with external instrument [34]. In [34] and [10], Machine Learning technique was utilized to characterize the microscopic process variations and related physical features. This section continues this thread, looking at how Machine Learning techniques are employed to characterize and configure the delay length of CCATDC. More specifically, the backward propagation of errors (backpropagation) algorithm is used. Backpropagation is a widely used technique in training artificial neural networks, and is often used with other optimization methods such as gradient descent. Usual backpropagation training consists of two phases: propagation and weight updating. Once an input vector is applied to the neural network model, it will be propagated from the input layer to the output layer. The output value of each input vector will be obtained and compared with the desired (golden) one; a loss function will be used to calculate the error for each neuron in the trained model, and update the weight parameters correspondingly.
Clustering and Classification
Published in Kayvan Najarian, Robert Splinter, Biomedical Signal and Image Processing, 2016
Kayvan Najarian, Robert Splinter
In this section, we explain backpropagation algorithm As in the perceptron, first, the weights of the network are initialized by randomly generated numbers. Backpropagation algorithm then updates the weights of the neural network through the propagation of error from the output backward to the neurons in the hidden and input layers. Backpropagation algorithm has four main stages as follows: (1) calculation of the output based on the current weights of network and the input patterns, (2) calculation of the error between the true target output and the predicted output, (3) backpropagation of the associated error to the previous layers, and finally, (4) adjustment of the weights according the backpropagated error. As can be seen, in steps 2 and 3 of backpropagation, each input units receives an input signal and transmits this input signal to hidden layer units. Then, each hidden layer unit computes its activation and sends it to the output units. For simplicity, from this point on, we assume that the network has only one hidden layer but the same algorithm can be easily extended to the networks for more than one hidden layer.
A deep neural network for valence-to-core X-ray emission spectroscopy
Published in Molecular Physics, 2023
The dropout regularisation technique (in which neurons are probabilistically ‘turned off’ and ‘dropped out’ of the weight update cycle with each feedforward/backpropagation epoch) is commonly used during optimisation to reduce model complexity and the propensity for overfitting [58]. The Monte-Carlo dropout technique (Figure 3(b)) additionally applies this probabilistic dropout at inference, or ‘prediction’ time, resulting in different outputs for the same input if the machine learning model is used repeatedly in inference mode. From N independent predictions with probabilistic dropout at inference time (analogous to sampling over N different DNN configurations), a mean prediction and standard deviation for each sample in the ‘held-out’ testing dataset can be derived. The latter, as in the ensembling approach (Section 2.3.1) quantifies a ‘dropout-configurational’ uncertainty which is related to a stochastic realisation of the Bayesian estimation for model uncertainty [59,60]. The effect of Monte-Carlo dropout quantifies the similarity of the input sample with the samples in the training dataset; the basis for this is that a higher or lower similarity will lead to a prediction which is less or more greatly affected by the dropout applied at inference time, respectively. A drawback of Monte-Carlo dropout is in its dependence on the hyperparametric dropout rate, p (Section 2.2), as the choice can have an effect on both on the accuracy of the machine learning model and the uncertainty estimation [58].
Soft computing paradigm for heat and mass transfer characteristics of nanofluid in magnetohydrodynamic (MHD) boundary layer over a vertical cone under the convective boundary condition
Published in International Journal of Modelling and Simulation, 2023
Hakeem Ullah, Muhammad Shoaib, Rafaqat Ali Khan, Kottakkaran Sooppy Nisar, Muhammad Asif Zahoor Raja, Saeed Islam
For eight scenarios of the HMT-CNF model, the convergence of mean square error, validation, and training developments are addressed in subfigures 6 and 7 (a, c, e, g) for test processes. With epochs 545, 278, 610, 448, 177, 208, 298, and 229 MSE nearby1.36E–09, 2.26E–09, 3.93E–10, 2.20E–09, 4.91E–09, 2.76E–09, 3.25 E-09 and 2.69E–09 gave the best network execution. The inferior the mean square error number the more exact and successful the approach’s execution is likely to be. Backpropagation calculates the gradient of the loss function in fitting a neural network, about the weights of the network for a single input-output example, and does so effectively, unlike a simple direct calculation of the gradient about every weight independently. Mu is the training gain and it stands for momentum constant or momentum parameter, which is involved in weight update appearance to get out of the problem of local minimum. The limit of mu is between 0 and 1. The approximates of Levenberg Marquardt gradient and step size Mu are closely [10–08, 10–08, 10–09, 10–08, 10–08, 10–08, 10–08, 10–08] and [9.99 × 10–08, 9.89 × 10–08, 9.98 × 10–08, 9.96 × 10–08, 9.95 × 10–08, 9.94 × 10–08, 9.91 × 10–08 and 9.93 × 10–08] had performed in subFigures 6 and 7 (b, d, f, h). The aforementioned results and graphical validations demonstrate that LMBT-NN is accurate, proficient, and convergent in every scenario of the HMT-CNF model.
Artificial neural networks to predict deformation modulus of rock masses considering overburden stress
Published in Geomechanics and Geoengineering, 2023
K. Tokgozoglu, C. H. Aladag, C. Gokceoglu
In Model 1, Em is explained by RMR and Ei values. Model 2 says that Em caused by RQD, UCS, WeatDeg, and Ei. Therefore, for Models 1 and 2, 2 and 4 neurons are, respectively, used in the input layers of ANN architectures. For each hidden layer, the number of neurons are changed from 1 to 20 in constructing the architectures of ANN. The data are divided into two sets namely training and test sets. ANN models are trained by utilising training set. Then, the prediction performances are evaluated over the test set. Backpropagation algorithm is used as training algorithm. It is a well-known fact that ANN approach is very good in modelling both nonlinear and linear structures by activation function (Egrioglu et al. 2015). In this study, in all neurons of the hidden layer, tangent sigmoid activation function, which is a nonlinear function, is used. Purelin linear activation function is utilised in the output neuron. Therefore, nonlinear and linear structures in the data can be modelled by ANN models. There are various activation functions used in the literature. Tangent sigmoid is used in this study since it is one of the most used functions in the literature (Gundogdu et al. 2016).