Abstract: Evaluations of a document are generated that indicate likelihoods of the document achieving its objectives. The evaluations are based on predictive characteristics of one or more outcomes of the client document that are indicative of whether the document will achieve its objectives. Specifically, a server receives the document from a client device. The server extracts a set of features from the document. The evaluations of the document are generated based on the predictive characteristics for the one or more outcomes of the document. The generated evaluations are provided to the client device such that the document can be optimized to achieve its desired objectives. The optimized document may also be sent to a posting server for posting to a computer network. The known outcomes of the optimized document are collected through reader responses to the document and analyzed to improve evaluations for other documents.

Abstract: A method and system of mitigating bias in a decision-making system are provided. A presence of bias is identified in one or more machine learning models. For each of the machine learning models, a presence of bias in an output of the model is determined. One or more options to mitigate a system bias during a processing stage, based on the identified presence of bias for each of the one or more models, are determined. One or more options to mitigate the system bias during a post-processing stage, based on the identified presence of bias in each output of the models, are determined. A combination of options is provided, including (i) a processing option for the processing stage, and (ii) a post-processing option for the post-processing stage, wherein the combination of options accommodates a threshold bias limit to the system bias and a total bias mitigation cost threshold.

Abstract: A method for machine-learning based process flow recommendation is provided. The method may include training a machine-learning model by at least processing training data with the machine-learning model. The training data may include a matrix representing one or more existing process flows by at least indicating actions that are performed on a document object to generate a subsequent document object. An indication that a first document object is created as part of a process flow may be received. In response to the indication, the trained machine-learning model may be applied to generate a recommendation to perform, as part of the process flow, an action to generate a second document object. Related systems and articles of manufacture, including computer program products, are also provided.

Abstract: Techniques for dynamic accuracy-based experimentation and deployment of machine learning (ML) models are described. Inference traffic flowing to ML models and the accuracy of the models is analyzed and used to ensure that better performing models are executed more often via model selection. A predictive component can evaluate which model is more likely to be accurate for certain input data elements. Ensemble techniques can combine inference results of multiple ML models to aim to achieve a better overall result than any individual model could on its own.

Abstract: An artificial intelligence-enabled device that handles delivery of user consumable information independent of network connectivity of the AI-enabled device, includes a memory and neural circuitry. The neural circuitry allocates a dedicated cache storage and determines a type of intelligent service on the AI-enabled device, for which first information is to be cached at the dedicated cache storage. The neural circuitry caches first information from a cloud server to a local sub-cache in the dedicated cache storage. The first information of the determined type of service is adaptively cached during at least one of a background activity or a foreground activity of the AI-enabled device, in accordance with an offline state or an online state of the AI-enabled device. The neural circuitry further controls delivery of user consumable information, based on a user input, on the AI-enabled device, based on the local sub-cache and supplemental information retrievable from the cloud server.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing large datasets using a computationally-efficient representation are disclosed. A request to apply a coverage algorithm to a large input dataset is received. The large dataset includes sets of elements. A computationally-efficient representation of the large dataset is generated by generating a reduced set of elements that contains fewer elements based on a defined probability. For each element in the reduced set, a determination is made regarding whether the element appears in more than a threshold number of sets. When the element appears in more than the threshold number, the element is removed from sets until the element appears in only the threshold number. The coverage algorithm is then applied to the computationally-efficient representation to identify a subset of the sets. The system provides data identifying the subset of the sets in response to the received request.

Abstract: An apparatus for learning a rank of an artificial neural network is configured to decompose a weight tensor into a first weight tensor and a second weight tensor. A set of rank selection parameters are applied to the first weight tensor and the second weight tensor to truncate the rank of the first weight tensor and the second weight tensor. The set of rank selection parameters are updated simultaneously with the weight tensors by averaging updates calculated for each rank selection parameter of the set of rank selection parameters.

Abstract: A method for model management includes receiving data on which to base a model, evaluating the received data against a plurality of existing models and data associated with each of the plurality of existing models, determining whether any of the plurality of existing models can be used as the model or as a basis to develop the model for the received data, and providing a user with the existing models that can be used as the model or as a basis to develop the model for the received data.

Abstract: Techniques facilitating simplified quantum programming are provided. In one example, a computer-implemented method comprises reducing, by a device operatively coupled to a processor, a first computing problem of a problem type to a second computing problem of the problem type, wherein the second computing problem is associated with a quantum circuit; facilitating, by the device, execution of the quantum circuit at a quantum computer, resulting in a first output corresponding to the second computing problem; and mapping, by the device, the first output to a second output corresponding to the first computing problem.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for rule generation and interaction. A rules engine is provided that does not require manual effort to generate or maintain high-quality rules over time for detecting malicious accounts or events. Rules no longer need to be manually added, tuned, or removed from the system. The system is able to determine the health of each rule and automatically add, tune, and remove rules to maintain a consistent, effective rule set.

Abstract: Aspects of the disclosure generally relate to computing devices and may be generally directed to devices, systems, methods, and/or applications for learning the operation of a computing device or software application, storing this knowledge in a knowledgebase, neural network, or other repository, and enabling autonomous operation of the computing device or software application with partial, minimal, or no user input.

Abstract: Technical solutions are described for storing weight in a crosspoint device of a resistive processing unit (RPU) array. An example method includes setting a state of each single bit counter from a set of single bit counters in the crosspoint device, the states of the single bit counters representing the weight to be stored at the crosspoint device. The method further includes adjusting electrical conductance of a resistor device of the crosspoint device. The resistor device includes a set of resistive circuits, each resistive circuit associated with a respective single bit counter from the set of single bit counters, the electrical conductance adjusted by activating or deactivating each resistive circuit according to a state of the associated single bit counter.

Abstract: A load control device may be configured to control multiple characteristics of one or more electrical loads such as the intensity and color of a lighting load. The load control device may switch from controlling one characteristic of the electrical loads to controlling another characteristic of the electrical loads based on the position and/or orientation of one or more components of the load control device. Such a position and/or orientation may be manipulated by moving the one or more components relative to a base portion of the load control device. The load control device may be a wall-mounted device or a battery-powered remote control device.

Abstract: Various embodiments describe user segmentation. In an example, potential rules are generated by applying a frequency-based analysis to user interaction data points. Each of the potential rules includes a set of attributes of the user interaction data points and indicates that these data points belong to a segment of interest. An objective function is used to select an optimal set of rules from the potential rules for the segment of interest. The potential rules are used as variable inputs to the objective function and this function is optimized based on interpretability and accuracy parameters. Each rule from the optimal set is associated with a group of the segment of interest. The user interaction data points are segments into the groups by matching attributes of these data points with the rules.

Abstract: A system and method of accelerating execution of a NN model, by at least one processor may include: receiving a first matrix A, representing elements of a kernel K of the NN model and a second matrix B, representing elements of an input I to kernel K; producing from matrix A, a group-sparse matrix A?, comprising G tensors of elements. The number of elements in each tensor is defined by, or equal to a number of entries in each index of an input tensor register used for a specific Single Instruction Multiple Data (SIMD) tensor operation, and all elements of A? outside said G tensors are null. The system and method may further include executing kernel K on input I, by performing at least one computation of the SIMD tensor operation, having as operands elements of a tensor of the G tensors and corresponding elements of the B matrix.

Abstract: A method for performing machine learning includes assigning processing jobs to a plurality of model learners, using a central parameter server. The processing jobs includes solving gradients based on a current set of parameters. As the results from the processing job are returned, the set of parameters is iterated. A degree of staleness of the solving of the second gradient is determined based on a difference between the set of parameters when the jobs are assigned and the set of parameters when the jobs are returned. The learning rates used to iterate the parameters based on the solved gradients are proportional to the determined degrees of staleness.

Abstract: Mechanisms are provided that generate a knowledge data structure for a medical condition. The mechanisms parse a natural language positional statement data structure representing a natural language positional statement corresponding to a medical condition, where the natural language positional statement specifies guidance information and patient attributes indicative of patients for which an action may be performed for the medical condition. The mechanisms extract the patient attributes and a grading value associated with the natural language positional statement. The mechanisms generate at least one weight value associated with each of the patient attributes based on the grading value.

Abstract: A machine learning system builds and uses computer models for identifying or predicting intensity of emotional reactions elicited by a particular video. Such computer models may also determine which particular emotional reaction corresponds to certain times during the video, and whether these reactions are positive or negative for a particular user. The computer models can also predict emotional reactions likely to be elicited by new videos based on learned correlations between video features and elicited emotional reactions.

Abstract: Computer systems and associated methods are disclosed to implement a model executor that dynamically selects machine learning models for choosing sequential actions. In embodiments, the model executor executes and updates an active model to choose sequential actions. The model executor periodically initiates a recent model and updates the recent model along with the active model based on recently chosen actions and results of the active model. The model executor periodically compares respective confidence sets of the two models' parameters. If the two confidence sets are sufficiently divergent, a replacement model is selected to replace the active model. In embodiments, the replacement model may be selected from a library of past models based on their similarity with the recent model. In embodiments, past models that exceed a certain age or have not been recently used as the active model are removed from the library.

Abstract: A system including a confidence assessment module that implements a neural network to assess the likelihood that codes associated with a patient's encounter with a healthcare organization are accurate. The confidence assessment module may also be incrementally trained.