A system and method for storing data in a distributed network having a plurality of datacenters distributed over a plurality of geographic regions. The method may involve receiving data, including metadata, uploaded to a first datacenter of the distributed network, receiving access information about previous data that was previously stored in the plurality of datacenters of the distributed network, predicting one or more of the plurality of geographic regions from which the uploaded data will be accessed based on the metadata and the access information, and instructing the uploaded data to be transferred from the first datacenter to one or more second datacenters located at each of the one or more predicted geographic regions.

A computer-implemented method, system, and non-transitory computer-readable storage medium for enhancing performance of a first model. The first model is trained with a training data set. A second model receives the training data set associated with the first model. The second model provides the first model with a hardness value associated with prediction of each data point of the training data set. The first model determines a confidence value regarding predicting each data point based on the training data set, and determines a ratio of the hardness value of a prediction of each data point by the second model with respect to the confidence value of the first model. The first model is retrained with a re-weighted training data set when the determined ratio is lower than a value of β.

Provided are processes of balancing between exploration and optimization with knowledge discovery processes applied to unstructured data with tight interrogation budgets. Natural language texts may be processed, such as into respective vectors, by a natural language processing model. An output vector of (or intermediate vector within) an example NLP model may include over 500 dimensions, and in many cases 700-800 dimensions. A process may manage and measure semantic coverage by defining geometric characteristics, such as size or a relative distance matrix, of a sematic space corresponding to an evaluation during which the natural language texts are obtained based on the vectors of the natural language texts. A system executing the process may generate a visualization of the semantic space, which may be reduced to or is a latent embedding space, by reducing the dimensionality of vectors while preserving their relative distances between the high and reduced dimensionality forms.

Methods, apparatus, systems, and articles of manufacture for training a neural network are disclosed. An example apparatus includes a training data segmenter to generate a partial set of labeled training data from a set of labeled training data. A matrix constructor is to create a design of experiments matrix identifying permutations of hyperparameters to be tested. A training controller is to cause a neural network trainer to train a neural network using a plurality of the permutations of hyperparameters in the design of experiments matrix and the partial set of labeled training data, and access results of the training corresponding of each of the permutations of hyperparameters. A result comparator is to select a permutation of hyperparameters based on the results, the training controller to instruct the neural network trainer to train the neural network using the selected permutation of hyperparameters and the labeled training data.

A system and method which allows the basic checkpoint-reverse-mode AD strategy (of recursively decomposing the computation to reduce storage requirements of reverse-mode AD) to be applied to arbitrary programs: not just programs consisting of loops, but programs with arbitrarily complex control flow. The method comprises (a) transforming the program into a formalism that allows convenient manipulation by formal tools, and (b) introducing a set of operators to allow computations to be decomposed by running them for a given period of time then pausing them, while treating the paused program as a value subject to manipulation.

A computer-implemented method of training machine learning models for making simulated estimations is provided. The method includes collecting, from a database, a set of historical data, applying one or more transformations to the set of historical data to create a set of model features, and separating the set of model features into one or more pools, each pool comprising one or more model features of the set that are homogeneous with respect to a common value. the method further includes, for each pool, dynamically creating a training set that includes the one or more sets of model features of the pool and at least some of the historical data. The method further includes, for each training set, training a machine learning model on the training set.

A non-transitory processor-readable medium stores instructions to be executed by a processor. The instructions cause the processor to receive a first trained machine learning model that generates a transcription based on a document. The instructions cause the processor to execute the first trained machine learning model and a second trained machine learning model to generate a refined transcription based on the transcription. The instructions cause the processor to execute a quality assurance program to generate a transcription score based on the document and the transcription. The instructions cause the processor to execute the quality assurance program to generate a refined transcription score based on the refined transcription and at least one of the document or the transcription. The at least one refined transcription score indicates an automation performance better than an automation performance for the at least one transcription score.

A search system performs a federated search across multiple databases and generates a ranked combined list of found genealogical records. The system receives a user query with one or more specified characteristics. The system may determine expanded characteristics derived from the specified characteristics. The system searches the various databases with the characteristics retrieving records according to the characteristics. The system combines the retrieved records and ranks them using a machine learning model. The machine learning model is configured to assign a weight to the records returned from each of the genealogical databases based on the characteristics specified in the user query. The machine learning model may be trained by any combination of one or more of: a Nelder-Mead method, a coordinate ascent method, and a simulated annealing method. The ranked combined results are provided in response to the user query.

Provided is a process including: obtaining, for a plurality of entities, entity logs, wherein: the entity logs comprise events involving the entities, a first subset of the events are actions by the entities, at least some of the actions by the entities are targeted actions, and the events are labeled according to an ontology of events having a plurality of event types; training, with one or more processors, based on the entity logs, a predictive machine learning model to predict whether an entity characterized by a set of inputs to the model will engage in a targeted action in a given geographic locale in the future; and storing the training the trained predictive machine learning model in memory.

A system maintains a knowledge layout to support the analysis of active events and determination of epicenter and aftershock nodes via an event reach stack. At an input layer of the event reach stack, the system may receive active event data. At a semantic layer, the system may parse the active event data to determine event phrases. Based on the event phrases, the system may identify epicenter nodes directly affected by the active event. At an analytic model layer, the system may successively determine aftershock nodes by traversing the knowledge layout outward from the epicenter nodes. The system then directs the response to the active event to the aftershock and epicenter nodes, via action at a focus response layer of the event reach stack.