China
Africa
Explore chapters and articles related to this topic
Introduction
Published in Peter Wlodarczak, Machine Learning and its Applications, 2019
It is often difficult to collect enough training data. Data is sometimes not publicly available or cannot be accessed for privacy or security reasons. To mitigate the problem of sparse data sets, synthetic data can be created, or training techniques such as cross-validation can be applied. Also, to train a learner, labeled data is needed. Raw data, such as emails, are not labeled, and producing labeled training sets can be a laborious task. Semi-supervised techniques such as active learning can be used in these situations. Active learning is a form of online learning in which the agent acts to acquire useful examples from which to learn [31]. In offline learning, all training data is available beforehand, whereas in online learning the training data arrive while the learner is trained. The learning algorithm can actively ask the agent to label data while it arrives. Data availability, whether labeled or not, is crucial for the success of a machine learning project and should be clarified before a machine learning project is initiated.
Neural-Network-Based AGC Design
Published in Hassan Bevrani, Takashi Hiyama, Intellyigent Automatic Generation Control, 2017
Hassan Bevrani, Takashi Hiyama
On the other hand, the learning process in an ANN could be offline or online. Offline learning is useful in feedforward applications such as classification and pattern recognition, while in feedback control applications, usually online learning, which is more complex, is needed. In an online learning process, the ANN must maintain the stability of a dynamical system while simultaneously learning and ensuring that its own internal states and weights remain bounded.
Machine learning for human movement understanding
Published in Advanced Robotics, 2020
Taizo Yoshikawa, Viktor Losing, Emel Demircan
Confusion matrix of offline learning is shown in Figure 9 and result of online learning is shown in Figure 10. In statistical classification, typically a supervised learning, to evaluate the accuracy of classification, it is common to create a confusion matrix and compare the number of correctly identified and the number of incorrectly identified ones. In the confusion matrix, each column of the confusion matrix represents the instances in an actual class and each row of the confusion matrix represents the instances in a predicted class. The diagonal elements of the confusion matrix represent the accuracy of the actual class (generated motion type input) as estimated by the classifier, and the other elements represent the error rate. Accuracy of offline classification (Figure 9) was 88−97% and online (Figure 10) was 72−93%. As expected, the accuracy of online learning was lower than offline learning. A major difference from offline learning is that when learning with new data, instead of recreating the model from scratch, the current model parameters are updated as needed. On the other hand, online learning is susceptible to noise such as outliers and susceptible to the latest data.
Performance comparison of different momentum techniques on deep reinforcement learning*
Published in Journal of Information and Telecommunication, 2018
In this study, multiple methods were used together in the training of the Othello game agent. The agent was trained by an online reinforcement learning methodology. Reinforcement learning methods may be used in an online or offline fashion. In online learning methodology, the structure is simultaneously trained while running on the problem. The last produced data is used in the training. On the contrary, a certain amount of data is generated without changing the structure in the offline learning approach and then this produced data is used to train the structure. Training in offline learning is similar to supervised learning. It is more challenging to train a structure with an online learning approach. If the training data is not sufficiently sparse, training may be stuck in local minima. Therefore, the training methodology and the structure must be carefully decided.