Abstract: Presented herein are techniques for training a central/global machine learning model in a distributed machine learning system. In the data sampling techniques, a subset of the data obtained at the local sites is intelligently selected for transfer to the central site for use in training the central machine learning model. In the model merging techniques, distributed local training occurs in each local site and copies of the local machine learning models are sent to the central site for aggregation of learning by merging of the models. As a result, in accordance with the examples presented herein, a central machine learning model can be trained based on various representations/transformations of data seen at the local machine learning models, including sampled selections of data-label pairs, intermediate representation of training errors, or synthetic data-label pairs generated by models trained at various local sites.

Abstract: A technology is provided for automated tuning of a machine learning model in a computing service environment. Predictive weights that include false positive outcomes, false negative outcomes, true positive outcomes, and true negative outcomes may be defined and/or received. A weight adjusted classification threshold, for use in a classification model of the machine learning model in a service provider environment, according to the predictive weights to enable the machine learning model to increase the total value of the machine learning model and decrease performance outcome errors. The improved classification threshold may be adjusted according to a change in the predictive weights. A data point may be classified according to the weight adjusted classification threshold in the classification model.

Abstract: Disclosed are methods, systems, and non-transitory computer-readable medium for executing neural network training for dynamically predicting apparel wearability. For example, a method may include generating a training data set comprising one or more historical data attributes of previously shipped apparel, training a neural network based on the training data set to configure one or more trained models to output a metric for any pair of a unique user identifier and a unique apparel identifier, storing one or more trained model objects, collecting prediction data comprising at least one prediction pair including a unique user identifier and a unique apparel identifier, predicting one or more predictive wearability metrics indicative of propensity to wear, dynamically generating one or more match pairs, and determining a match wearability metric for each of the one or more match pairs based on the predicted one or more predictive wearability metrics.

Abstract: Technologies for artificial neural network training include a computing node with a host fabric interface that sends a message that includes one or more artificial neural network training algorithm values to another computing node in response to receipt of a request to send the message. Prior to sending the message, the host fabric interface may receive a request to quantize the message and quantize the message based on a quantization level included in the request to generate a quantized message. The quantization message includes one or more quantized values such that each quantized value has a lower precision than a corresponding artificial neural network training algorithm value. The host fabric interface then transmits the quantized message, which includes metadata indicative of the quantization level, to another computing node in response to quantization of the message for artificial neural network training. Other embodiments are described and claimed.

Abstract: A learning model difference providing method that causes a computer to execute a process which includes: calculating a mismatch degree between prediction data about arbitrary data included in a plurality of pieces of data that are input by using an application program, the prediction data being obtained by the plurality of pieces of data and a learning model in accordance with a purpose of use of the application program, and data that are specified for the arbitrary data; assessing whether or not the calculated mismatch degree exceeds a first degree; and transmitting the mismatch degree to a providing source of the learning model in a case where the mismatch degree is assessed as exceeding the first degree.

Abstract: A device, system, and method for training or prediction of a neural network. A current value may be stored for each of a plurality of synapses or filters in the neural network. A historical metric of activity may be independently determined for each individual or group of the synapses or filters during one or more past iterations. A plurality of partial activations of the neural network may be iteratively executed. Each partial-activation iteration may activate a subset of the plurality of synapses or filters in the neural network. Each individual or group of synapses or filters may be activated in a portion of a total number of iterations proportional to the historical metric of activity independently determined for that individual or group of synapses or filters. Training or prediction of the neural network may be performed based on the plurality of partial activations of the neural network.

Abstract: Disclosed herein are cloud-based machine learning systems and methods for monitoring networked devices to identify and classify characteristics, to infer typical or atypical behavior and assign reputation profiles across various networked devices, and to make remediation recommendations. In some embodiments, a cloud-based machine learning system may learn the typical operation and interfacing of a plurality of reputable devices that are known to be free from malicious software and other threats. In some embodiments, a cloud-based machine learning system may learn the typical operation and interfacing of a device, and may identify atypical operations or interfaces associated with that device by comparing the operations and interfaces to those of a plurality of networked devices or to those of a defined standard reference device.

Abstract: A computer system can perform a semi-supervised machine learning processes to cluster a plurality of entities within a population based on their features and associated labels. The computer system can generate visualization data representing the clusters of entities and associated labels for displaying on a user interface. A user can review the clustering of entities and use the user interface to add or modify the labels associated with a particular entity or set of entities. The computer system can use the user's feedback to update the labels and then re-determine the clustering of entities using the semi-supervised machine learning process with the updated labels as input. As such, the computer system can use the user's feedback to improve the accuracy of the machine learning model without requiring a larger amount of labeled input data.

Abstract: Mechanisms are provided for generating a dashboard recommendation based on tracked user input patterns and the operation of predictive analytics. The mechanisms present a dashboard interface to a user via a client computing device, and track user inputs to the client computing device at least during and after presentation of the dashboard interface to the user via the client computing device. The mechanisms apply predictive analytics to the tracked user inputs to predict a type of data the user is attempting to access, and correlate the predicted type of data with one or more portions of one or more other dashboard interfaces that provide a representation of data having a type matching the predicted type of data. The mechanisms output a recommendation output to the user via the client computing device recommending the user access the one or more other dashboard interfaces.

Abstract: Disclosed are methods, systems, and non-transitory computer-readable medium for assigning wearable items in a subscription electronics transactions platform. For example, a method may include: generating a grid based on information regarding historically shipped wearable items, wherein the grid comprises at least a first cell and a second cell; determining an average percentage indicating how many wearable items have been used and an average predictive wearability metric for wearable items indicative of a propensity of a user to use the wearable items per number of wearable items shipped for each cell; generating a mapping configured to convert a predictive wearability metric to a squashed predictive wearability metric; and converting a first predictive wearability metric to a first squashed wearability metric based on the generated mapping.

Abstract: Systems, apparatuses and methods may provide for technology that assigns confidence levels to data bins containing similarity data and length of stay data, wherein the similarity data and the length of stay data correspond to a plurality of previous admissions. Additionally, the confidence levels may be weighted based on a distribution metric that assigns higher weights to denser regions. A length of stay of a target admission may be predicted based on the weighted confidence levels.

Abstract: Disclosed is a novel system, and method to evaluate a prediction of a possibly unknown outcome out of a plurality of predictions of that outcome. The method begins with accessing a particular prediction of an outcome out of a plurality of predictions of that outcome in which the outcome may be unknown. Next, a subsample of the plurality of predictions of the outcome is accessed. The subsample can possibly include the particular prediction. A consensus prediction of the outcome based on the subsample of the plurality of predictions is determined. A proximity of the particular prediction to the consensus prediction is determined Each prediction is ranked out of the plurality of predictions in an order of a closest in proximity to the consensus prediction to a farthest in proximity to the consensus prediction.

Abstract: A learning graph is generated for documents according to a sequencing approach. The learning graph includes nodes corresponding to the documents and edges. Each edge connects two of the nodes and indicates a sequencing relationship between two of the documents to which the two of the nodes correspond that specifies an order in which the two of the documents are to be reviewed in satisfaction of the learning goal. The learning graph is a directed graph specifying a learning path through the documents to achieve a learning goal in relation to a subject.

Abstract: Methods, systems and apparatus, including computer programs encoded on computer storage media, for training a value neural network that is configured to receive an observation characterizing a state of an environment being interacted with by an agent and to process the observation in accordance with parameters of the value neural network to generate a value score. One of the systems performs operations that include training a supervised learning policy neural network; initializing initial values of parameters of a reinforcement learning policy neural network having a same architecture as the supervised learning policy network to the trained values of the parameters of the supervised learning policy neural network; training the reinforcement learning policy neural network on second training data; and training the value neural network to generate a value score for the state of the environment that represents a predicted long-term reward resulting from the environment being in the state.

Abstract: Techniques for reinforcement for bots are described. In one embodiment, an apparatus may comprise a bot application interface component operative to receive a plurality of bot capability catalogs for a plurality of bots at a bot-service system; a client communication component operative to receive a plurality of user service prompts from a plurality of user client devices; and an interaction processing component operative to determine the selected bots of the plurality of bots for each of the plurality of user prompts by matching the plurality of user service prompts against the plurality of bot capability catalogs using a bot capability table generated by a natural-language machine-learning component; record a bot interaction history based on user interactions with the selected bots; and update the natural-language machine-learning component based on the bot interaction history. Other embodiments are described and claimed.

Abstract: An approach is provided to receive, at a question answering (QA) system, a question and identify a politeness corresponding to a number of terms corresponding to the question that are included in a corpus of the QA system. The approach identifies the politeness of one or more terms included in each of a set of candidate answers responsive to the question. Finally, the approach scores each of the candidate answers, with the scoring being based, in part, on the politeness identified for each of the terms.

Abstract: Embodiments are described for minimizing a wait time for a rider after sending a ride request for a vehicle. An example computer-implemented method includes receiving a ride request, the request being for travel from a starting location to a zone in a geographic region during a specified timeslot. The method further includes predicting travel demand based on a number of ride requests in the zone during the specified timeslot. The method further includes requesting transport of one or more vehicles to the zone in response to the predicted number of ride requests when the travel demand is predicted to exceed a number of vehicles in the zone during the specified timeslot.

Abstract: Attribute data structures can be updated to indicate joint relationships among attributes and predictive outputs in training data that can be used for training automated modeling system. A data structure that stores training data for training an automated modeling algorithm can be accessed. The training data can include first data for a first attribute and second data for a second attribute. The data structure can be modified to include a derived attribute that indicates a joint relationship among the first attribute, the second attribute, and a predictive output variable. The automated modeling algorithm can be trained with the first attribute, the second attribute, and the derived attribute.

Abstract: Embodiments are described for minimizing a wait time for a rider after sending a ride request for a vehicle. An example computer-implemented method includes receiving a ride request, the request being for travel from a starting location to a zone in a geographic region during a specified timeslot. The method further includes predicting travel demand based on a number of ride requests in the zone during the specified timeslot. The method further includes requesting transport of one or more vehicles to the zone in response to the predicted number of ride requests when the travel demand is predicted to exceed a number of vehicles in the zone during the specified timeslot.

Abstract: In a machine learning environment, missing values can be imputed based upon an expectation maximization style approach. A system partitions a dataset, and uses a first partition as a training subset and a second partition as a verification subset. The training subset is used to train a machine learning model, which is then used to impute missing values in the second subset. The subsets may be swapped and the process iterates to predict missing values in the dataset with a high degree of accuracy, thereby improving both the accuracy of the machine learning model and the accuracy of the imputed values. The noise in the value prediction is reduced through a linear regression setting to account for heterogeneity in the dataset.