China
Africa
Explore chapters and articles related to this topic
Data fusion and machine learning for bridge damage detection
Published in Joan-Ramon Casas, Dan M. Frangopol, Jose Turmo, Bridge Safety, Maintenance, Management, Life-Cycle, Resilience and Sustainability, 2022
Hao Wang, Giorgio Barone, Alister Smith
The proposed convolutional AE employs 12 1D convolutional layers, each having a dropout rate of 0.2, that is, 20% of neurons are randomly selected and ignored during the training process. An early-stopping strategy was employed to terminate the training process when performance on a validation dataset starts to degrade. Both dropout and early stopping avoided overfitting the model. The Nadam (Nesterov-accelerated Adaptive Moment Estimation) optimiser was used with an initial learning rate of 0.001, which reduces as the training epoch increases. The proposed convolutional AE was implemented with the Keras v2.7.0 and Python 3.8.5, and the calculations were performed by GPU computing on the NVIDIA Quadro P1000 graphics card to accelerate the training process. The details of the proposed convolutional AE are provided in Table 2.
Ground Truth Data for Remote Sensing Image Classification
Published in Anil Kumar, Priyadarshi Upadhyay, A. Senthil Kumar, Fuzzy Machine Learning Algorithms for Remote Sensing Image Classification, 2020
Anil Kumar, Priyadarshi Upadhyay, A. Senthil Kumar
Once the classification algorithm has been trained (called a fitted model), it can be used to classify the data called a validation dataset (James, 2013). From a validation dataset, an unbiased evaluation of a model fit is done while tuning of parameters has been done using a training dataset (Brownlee, 2017) [e.g. the number of hidden units in a neural network (Ripley, 1996)]. Validation datasets can be used to regularize by early stopping. The early stopping means stop training when the error on the validation dataset increases, as this is a sign of overfitting to the training dataset (Prechelt and Genevieve, 2012). It has been noticed that a validation dataset’s error fluctuates during training, generating multiple local minima. Due to this, it has become difficult and necessary to make complicated rules to know when overfitting has truly begun (Prechelt and Genevieve, 2012). Later, the test dataset is used to provide an unbiased evaluation of the trained classified model. If the test data has never been used in training, then the test dataset is also called a holdout dataset.
Hybrid Model Identification and Discrimination with Practical Examples from the Chemical Industry
Published in Jarka Glassey, Moritz von Stosch, Hybrid Modeling in Process Industries, 2018
Andreas Schuppert, Thomas Mrziglod
Usually early stopping is used as a criterion to terminate the numerical iteration and to avoid overfitting. By that means a split into a training and a validation set of the available data is used. The training set is then used to minimize Equation 4.8 with a suitable previously mentioned optimization method. The prediction error of the model on the validation set is considered to control the model behavior on unknown data. For example, the iteration procedure could be terminated if the model error on the validation set increases. As an alternative, a maximum number of iterations is performed, and the parameter values corresponding to the smallest validation set error are returned as the best solution.
TW3-based ROI detection and classification using a chaotic ANN and DNN-EVGO architecture for an automated bone age assessment on hand X-ray images
Published in The Imaging Science Journal, 2023
Thangam Palaniswamy, Mahendiran Vellingiri, M. Ramkumar Raja
In deep learning, the accuracy of a model's predictions is typically measured using a validation set, which is a subset of the data that is held out from the training process. The validation set is used to evaluate the model's performance on new, unseen data. The number of epochs in the training process is one of the hyper-parameters that can be tuned to improve the model's accuracy. An epoch is a single pass through the entire training set, and increasing the number of epochs allows the model to see the training data more times and potentially learn more complex patterns. Up to a certain point, increasing the count of epochs can enhance the model's accuracy on both the training and validation sets. However, after a certain number of epochs, the model may start to overfit the training data, meaning it memorizes the training set and performs unwell on new, unobserved data. The ideal count of epochs is depending on a particular dataset and model architecture. In practice, it is common to use early stopping to avoid overfitting and choose the count of epochs that result in the best performance on a set of verification. Increasing the count of epochs can improve the accuracy of a model's predictions up to a certain point, but the optimal number of epochs depends on the specific dataset and model architecture, and overfitting can occur if the number of epochs is very high.
Identifying heart disease risk factors from electronic health records using an ensemble of deep learning method
Published in IISE Transactions on Healthcare Systems Engineering, 2023
Linkai Luo, Yue Wang, Daniel Y. Mo
The model was trained in multiple epochs. It was tested on the validation set after each training epoch. To avoid model overfitting, an early stopping technique is used (Girosi et al., 1995), in which the model training terminates when the training loss on the validation set stops decreasing. The final training results are based on the updated parameters in the current epoch. The next training epoch is conducted if the loss value continues to decrease. During the experiment, we noticed that while overall performance tends to improve with the training epochs (before the stopping epoch), the performance of a single class may not. For example, the overall risk identification performance in the 8th training epoch is better than the 7th epoch. However, the individual risk factor of CAD classification in the 8th epoch is deteriorated. This suggests that the model parameters in the 7th epoch are better for identifying CAD risk factors. As a result, overall performance improves at the expense of certain classes. This may not be ideal for identifying risk factors for heart disease.
Satellite and instrument entity recognition using a pre-trained language model with distant supervision
Published in International Journal of Digital Earth, 2022
Ming Lin, Meng Jin, Yufu Liu, Yuqi Bai
The number of parameters of the pre-trained language model enables it to fit the named entity recognition task of this paper. The distantly supervised training data suffer from the mislabeling and omission of named entities. The features of these false labels are harder to learn and require more iterations to fit. Early stopping is an important strategy to avoid overfitting. At first, the model has only semantic and syntactic knowledge learned from a large text corpus. Then we fine-tune the model on the training data. Training stops after iterating to a specified threshold. This prevents overfitting of the incomplete annotated labels while retaining knowledge learned during pre-training. Figure 5 shows the model’s ability to fit the data in the stages of underfitting, early stopping, and overfitting. Early stopping effectively improves generalization performance.