Transfer (computing)

Abstract: Transfer of pre-trained representations improves sample efficiency and simplifies hyperparameter tuning when training deep neural networks for vision. We revisit the paradigm of pre-training on large supervised datasets and fine-tuning the model on a target task. We scale up pre-training, and propose a simple recipe that we call Big Transfer (BiT). By combining a few carefully selected components, and transferring using a simple heuristic, we achieve strong performance on over 20 datasets. BiT performs well across a surprisingly wide range of data regimes—from 1 example per class to 1M total examples. BiT achieves 87.5% top-1 accuracy on ILSVRC-2012, 99.4% on CIFAR-10, and 76.3% on the 19 task Visual Task Adaptation Benchmark (VTAB). On small datasets, BiT attains 76.8% on ILSVRC-2012 with 10 examples per class, and 97.0% on CIFAR-10 with 10 examples per class. We conduct detailed analysis of the main components that lead to high transfer performance.

Abstract: Transfer of pre-trained representations improves sample efficiency and simplifies hyperparameter tuning when training deep neural networks for vision. We revisit the paradigm of pre-training on large supervised datasets and fine-tuning the model on a target task. We scale up pre-training, and propose a simple recipe that we call Big Transfer (BiT). By combining a few carefully selected components, and transferring using a simple heuristic, we achieve strong performance on over 20 datasets. BiT performs well across a surprisingly wide range of data regimes -- from 1 example per class to 1M total examples. BiT achieves 87.5% top-1 accuracy on ILSVRC-2012, 99.4% on CIFAR-10, and 76.3% on the 19 task Visual Task Adaptation Benchmark (VTAB). On small datasets, BiT attains 76.8% on ILSVRC-2012 with 10 examples per class, and 97.0% on CIFAR-10 with 10 examples per class. We conduct detailed analysis of the main components that lead to high transfer performance.

Abstract: A data network with a remote data facility for providing redundant data storage and for enabling concurrent point-in-time backup operations. A local data processing system with a data facility stores a data base and processes applications. A second system, physically separated from the first system, includes a data facility that normally mirrors the data in the first system. In a backup mode, the second system is enabled to transfer data from its data facility to a backup facility concurrently with, but independently of, the operation of the first system. On completion of the backup operation, the second system reconnects with and synchronizes with the first system thereby to reestablish the mirroring operation of the second system.

Abstract: A network computer system providing for the fault tolerant storage and retrieval of data files includes a client system connected to a data communication network that may source a first data transfer request to said data communication network for the transfer or retrieval of data. A first server system, including first medium for storing data files, is connected to the data communication network so as to be responsive to first data transfer requests. A second server system, including second medium for storing data files is also connected to said data communication network to also be responsive to first data transfer requests. A control protocol, established between the first and second server systems, coordinates an asymmetric response by the first and second server systems to a first data transfer request, such that file data transferred by the client with the first data transfer request is replicated to the first and second storing mediums and such that file data transferred to the client system in response to the first data transfer is non-replicatively provided to the client system by either the first or second server system.