Abstract
In this paper, a novel algorithm is proposed for inferring online learning tasks efficiently. By a carefully designed scheme, the online learning problem is first formulated as a state feedback control problem for a series of finite-dimensional systems. Then, the online linear quadratic regulator (OLQR) learning algorithm is developed to obtain the optimal parameter updating. Solid mathematical analysis on the convergence and rationality of our method is also provided. Compared with the conventional learning methods, our learning framework represents a completely different approach with optimal control techniques, but does not introduce any assumption on the characteristics of noise or learning rate. The proposed method not only guarantees the fast and robust convergence but also achieves better performance in learning efficiency and accuracy, especially for the data streams with complex noise disturbances. In addition, under the proposed framework, new robust algorithms can be potentially developed for various machine learning tasks by using the powerful optimal control techniques. Numerical results on benchmark datasets and practical applications confirm the advantages of our new method.
Highlights
As an important subtopic of machine learning, online learning has attracted increasing attention during the past decade due to its extensive applications to realistic modeling problems, for instance, online advertising, financial quantitative transaction, and mechanical damage detection [1]–[4]
We introduce the state feedback control theory into the modeling of data streams and propose a novel online learning approach
Since a completely different approach is explored in our framework, there is no need to introduce any online adjustment to the learning parameters, complex data window or pruning techniques, which enables our method to overcome the aforementioned limitations of the existing methods
Summary
As an important subtopic of machine learning, online learning has attracted increasing attention during the past decade due to its extensive applications to realistic modeling problems, for instance, online advertising, financial quantitative transaction, and mechanical damage detection [1]–[4]. Another major approach is the online version of batch learning algorithms [11]–[13] In these algorithms, the learning model is updated by repeatedly solving the corresponding regularized error minimization problem when the new instances are subsequently added. The updates will be continuously misguided, and the convergence rate will be seriously affected Another limitation is that some key learning parameters, such as the length of the window in the moving window regression [26] and the learning rate in the gradient based methods [27], are difficult to be adjusted. Since a completely different approach is explored in our framework, there is no need to introduce any online adjustment to the learning parameters, complex data window or pruning techniques, which enables our method to overcome the aforementioned limitations of the existing methods. I denotes the identity matrix, and 0 ≺ S ≺ I means that the symmetric matrix S and I − S are both positive definite
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
