Abstract: A computation node of a neural network training system is described. The node has a memory storing a plurality of gradients of a loss function of the neural network and an encoder. The encoder encodes the plurality of gradients by setting individual ones of the gradients either to zero or to a quantization level according to a probability related to at least the magnitude of the individual gradient. The node has a processor which sends the encoded plurality of gradients to one or more other computation nodes of the neural network training system over a communications network.

Abstract: Systems and methods for importing data from data sources over a network while correcting and transforming the data are described. A data migration server receives data from various data sources. The data migration server repairs and transforms the received data before transmitting the data to a new data server. In some embodiments, the data migration server repairs data before formatting the data, and in other embodiments, the data migration server formats data before repairing the data. The data migration server also verifies the accuracy of the data, either by comparing the data against data requested and received from third-party servers or in reference to other data from the data sources.

Abstract: Systems and methods for automatically identifying data dependencies for reports are described. In one embodiment, a method includes: instructing a first reporting utility to generate a first report according to a set of parameters, the first report based on data stored in a database; modifying, directly or indirectly, at least one data element in the database; instructing the first reporting utility to generate a second report according to the set of parameters, the second report based on data stored in the database; comparing the first report to the second report; and identifying, based on the comparison, at least one field within the database used by the first reporting utility to generate the first and second reports.

Abstract: Systems and methods for importing data from data sources over a network while correcting and transforming the data are described. A data migration server receives data from various data sources. The data migration server repairs and transforms the received data before transmitting the data to a new data server. In some embodiments, the data migration server repairs data before formatting the data, and in other embodiments, the data migration server formats data before repairing the data. The data migration server also verifies the accuracy of the data, either by comparing the data against data requested and received from third-party servers or in reference to other data from the data sources.

Abstract: Systems and methods for automatically identifying data dependencies for reports are described. In one embodiment, a method includes: instructing a first reporting utility to generate a first report according to a set of parameters, the first report based on data stored in a database; modifying, directly or indirectly, at least one data element in the database; instructing the first reporting utility to generate a second report according to the set of parameters, the second report based on data stored in the database; comparing the first report to the second report; and identifying, based on the comparison, at least one field within the database used by the first reporting utility to generate the first and second reports.

Abstract: Systems and methods for automatic quality assurance of reports generated by workflow software may include: instructing a first reporting utility to generate a first report according to a set of parameters, the first report based on data stored in a database; instructing a second reporting utility to generate a second report according to the set of parameters, the second report based on data stored in the database; comparing the first report to the second report; and identifying at least one difference between the first and second report. The method may also include producing appropriate output identifying the differences. The method may also include determining some differences are solely formatting differences between the reports, and ignoring or outputting those differences in a different manner.