China
Africa
Explore chapters and articles related to this topic
Overview of Deep Learning Algorithms Applied to Medical Images
Published in Ayman El-Baz, Jasjit S. Suri, Big Data in Multimodal Medical Imaging, 2019
Behnaz Abdollahi, Ayman El-Baz, Hermann B. Frieboes
Machine learning algorithms require input data to learn and approximate mathematical functions, as these algorithms are trained to learn the correlation of the input data to the corresponding output. The desired function is thus trained and evaluated on the provided dataset to make it generalizable on unknown data. For optimal performance, the design, development, and evaluation of these algorithms require a large amount of training data. According to specific objectives, machine learning methods include supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, transfer learning, and deep learning, which is based on artificial neural networks (1,2). Generally machine learning models are designed and developed on 70–80% of the collected data, which is defined as the training dataset. The developed model is evaluated on a smaller subset of the dataset, which was not part of the training set and is called the validation set. The model’s performance is measured on the validation set, and the hyperparameters are fine tuned. Lastly, the tuned model is tested on another subset of the dataset called the test set. The purpose of the test set is to evaluate the generalizability of the tuned model on unknown data.
Classification of Text Data in Healthcare Systems – A Comparative Study
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
The second step is the application of pre-processing techniques, if required. It shall be noted that different processing techniques are efficient for different data sets. Hence, the processing step shall be applied iteratively on the data set to evaluate the effect of each particular pre-processing technique. Then the hyperparameter selection step takes place. Hyperparameters are the parameters that are not learned by the algorithms, they shall be tuned iteratively as well as parameters of the classifiers iteratively. Then the fine-tuning step takes place. In this phase, selected parameters are tuned with the use of input data. The result of the provided methodology is the feasible model with tuned hyperparameters for the selected data set.
Ideation
Published in Walter R. Paczkowski, Deep Data Analytics for New Product Development, 2020
Some have argued that LDA is not a good method because it is inconsistent. A different answer results from the same data set each time it is used. In addition, it uses hyperparameters so more “tuning” is required to get the right set of hyperparameters. Hyperparameters are those parameters that are set prior to model estimation in contrast to the parameters that are estimated. Hyperparameters define the model and its complexity. See Alpaydin [2014]. The method, however, is improving at a rapid speed as research continues to add to its foundations.10
Wind field forecasting using a novel method based on convolutional neural networks and bidirectional LSTM
Published in Ships and Offshore Structures, 2023
In order to find the best hyper parameters of the system (layer size, optimizer, etc.), the grid search method is utilised. Grid search is a hyperparameter tuning method that involves defining a set of hyperparameters and their possible values, and then systematically searching over all possible combinations of these hyperparameters to find the combination that yields the best performance on a validation set. To perform grid search, the practitioner defines a grid of hyperparameters, where each hyperparameter is assigned a set of possible values. For the proposed model, the hyperparameters include the size of layers, number of layers, optimizer type, and number of epochs. The practitioner might define a grid with, say, 5 possible values for number of layers (e.g. 3, 5, and 7) and two possible values for size of layers (e.g. 20 and 50). The grid would then consist of six possible combinations of hyperparameters. The grid search method then trains and evaluates a model for each combination of hyperparameters on a validation set. The performance metric is recorded for each combination, and the combination with the best performance is selected as the optimal hyperparameters. Grid search has the advantage of being simple and easy to implement, and it guarantees that the optimal hyperparameters will be found if the grid is defined appropriately. However, it can be computationally expensive, especially when dealing with a large number of hyperparameters and their possible values.
Evaluation and prediction of punctuality of vessel arrival at port: a case study of Hong Kong
Published in Maritime Policy & Management, 2023
Zhong Chu, Ran Yan, Shuaian Wang
The value of ‘none’ value for max_depth in Table 9 means that there is no limitation on the depth of the tree, i.e. the nodes can be expanded until there is only one sample in each leaf node or all of the leaves are pure (i.e. with samples that have the same output). We implement a grid search and the K-fold cross-validation (K-fold CV) method to find the best values for the hyperparameters shown in Table 9. A grid search is a tuning method that exhaustively searches combinations of the candidate values for all candidate hyperparameters to find the set of hyperparameter values that leads to the best performance on the validation set(s) (Breiman 2001; Arlot and Celisse 2010). The search ranges and intervals of the hyperparameters for the RF model are listed in Table 10.
Optimized Convolutional Neural Network-Based Adaptive Controller for 6-Phase DFIG-Based Wind Energy Systems
Published in Electric Power Components and Systems, 2023
Arunkumar Azhakappan, Agees Kumar Chellappan, Murugan Sethuramalingam
A development of CNN is the deep convolutional neural network (DCNN) is proposed for the six-phase DIFG WECS problem. It features hierarchical patch-based convolution operations, which are often used in data analysis, pattern recognition, feature extraction, and the identification of pictures and videos. By abstracting pictures on several feature levels, DCNN, which was derived from the conventional ANN, lowers the computing cost. The use of unprocessed raw input data has been guided by DCNN's superior performance. The convolutional and fully connected layers’ weights and biases are adjusted using the gradient descent technique. The efficiency of the DCNN model is significantly influenced by a variety of hyperparameters, including the learning rate, momentum, validation frequency, L2R regularization, number of epochs, and activation function. A hyperparameter is a parameter that is employed to control the training process in machine learning by setting its value before the commencement of training. It is distinct from model parameters, which are learned by the algorithm during the training process. To ensure that DCNN performs at its best, this work focuses on optimizing these hyperparameters. Even though checking each parameter increases the computing complexity, these factors are crucial for developing an optimal solution. In this study, FPA has been selected for hyperparameter optimization due to its speed of convergence and few control parameters. Figure 2 depicts the flow diagram for the hyperparameter optimization of DCNN using FPA.