No abstract

This article focuses on gradient-based backpropagation algorithms that use either a common adaptive learning rate for all weights or an individual adaptive learning rate for each weight and apply the Goldstein/Armijo line search. The learning-rate adaptation is based on descent techniques and estimates of the local Lipschitz constant that are obtained without additional error function and gradient evaluations. The proposed algorithms improve the backpropagation training in terms of both convergence rate and convergence characteristics, such as stable learning and robustness to oscillations. Simulations are conducted to compare and evaluate the convergence behavior of these gradient-based training algorithms with several popular training methods.

In this paper the development, convergence theory and numerical testing of a class of gradient unconstrained minimization algorithms with adaptive stepsize are presented. The proposed class comprises four algorithms: the ÿrst two incorporate techniques for the adaptation of a common stepsize for all coordinate directions and the other two allow an individual adaptive stepsize along each coordinate direction. All the algorithms are computationally e cient and possess interesting convergence properties utilizing estimates of the Lipschitz constant that are obtained without additional function or gradient evaluations. The algorithms have been implemented and tested on some well-known test cases as well as on real-life artiÿcial neural network applications and the results have been very satisfactory.

There is a pressing need for the archiving and curation of raw X-ray diffraction data. This information is critical for validation, methods development and improvement of archived structures. However, the relatively large size of these data sets has presented challenges for storage in a single worldwide repository such as the Protein Data Bank archive. This problem can be avoided by using a federated approach, where each institution utilizes its institutional repository for storage, with a discovery service overlaid. Institutional repositories are relatively stable and adequately funded, ensuring persistence. Here, a simple repository solution is described, utilizing Fedora open-source database software and data-annotation and deposition tools that can be deployed at any site cheaply and easily. Data sets and associated metadata from federated repositories are given a unique and persistent handle, providing a simple mechanism for search and retrieval via web interfaces. In addition to ensuring that valuable data is not lost, the provision of raw data has several uses for the crystallographic community. Most importantly, structure determination can only be truly repeated or verified when the raw data are available. Moreover, the availability of raw data is extremely useful for the development of improved methods of image analysis and data processing.

The Store.Synchrotron service, a fully functional, cloud computing-based solution to raw X-ray data archiving and dissemination at the Australian Synchrotron, is described. The service automatically receives and archives raw diffraction data, related metadata and preliminary results of automated data-processing workflows. Data are able to be shared with collaborators and opened to the public. In the nine months since its deployment in August 2013, the service has handled over 22.4 TB of raw data (∼1.7 million diffraction images). Several real examples from the Australian crystallographic community are described that illustrate the advantages of the approach, which include real-time online data access and fully redundant, secure storage. Discoveries in biological sciences increasingly require multidisciplinary approaches. With this in mind, Store.Synchrotron has been developed as a component within a greater service that can combine data from other instruments at the Australian Synchrotron, as well as instruments at the Australian neutron source ANSTO. It is therefore envisaged that this will serve as a model implementation of raw data archiving and dissemination within the structural biology research community.