Abstract: The present disclosure relates to the technical field of semiconductor integrated circuits and discloses a multi-mode array structure for in-memory computing, and a chip, including: an array of memory cells, function lines corresponding to all the memory cells measured by rows in the array of memory cells, and complementary function lines and bit lines BL corresponding to all the memory cells measured by columns in the array of memory cells. According to the present disclosure, the TCAM function and CNN and SNN operations are enabled; the multi-mode array for in-memory computing herein goes beyond the limits of the von Neumann architecture by integrating the multiple modes of storage and computation, achieving efficient operation and computation; in addition to solving the computing power problem, a new array mode is provided to promote the development of high-integration circuits.

Abstract: A computer-implemented method of predicting energy use for a route including inputting map data of roads included in K trips in a geographical area, predictors of rate of energy use along the roads, and energy consumption data of the K trips. The method includes dividing each of the roads in the map data for all the trips into segments of length measure ?i; grouping the segments from the trips into a number N of clusters, using an algorithm to build a model predicting the weights Wj based on solving a system of equations, one per trip, assigning the predicted weight applied to the cluster in which the segment was grouped and storing a segment ID with the corresponding cluster ID or predicted rate of energy use Yi to allow prediction of energy use for a route in the geographical area incorporating one or more of the segments.

Abstract: This disclosure describes a method to verify the design of an article of manufacture that includes a software program executing on a workstation that executes the following steps: providing the shape of a first component and providing the shape of a second component; providing a boundary distance constraint; constructing a shape spectrum of the exterior boundary surface of closest approach associated with the shape of the first component, the shape of the second component, and the boundary distance constraint; verifying that the arrangement of the shape of the first component and the shape of the second component satisfy the boundary distance constraint; evaluating a subderivative of the shape spectrum of the exterior boundary surface of closest approach associated with the shape of the first component, the shape of the second component; where verifying that the boundary distance constraint is satisfied between the shape of the first component and the shape of the second component and when satisfied, the desi

Abstract: Disclosed embodiments include methods and systems for classifying test data. In one embodiment a method includes determining one or more variable types in a multivariate test vector within a data set, and for a plurality of machine-learning models, determining a closest match between variable types used by (to train) the machine-learning models and the determined variable types for the test vector. In response to determining a closest match for one machine-learning model, a corresponding machine-learning model is selected and the test vector is classified using the selected model. In response to determining a closest match for multiple machine-learning models, a similarity is determined between a probability distribution for the test data set and the probability distributions for the multiple machine-learning models to generate similarity values for each of the models.

Abstract: This specification describes methods and systems for accelerating attribute data access for graph neural network (GNN) processing. An example method includes: receiving a root node identifier corresponding to a node in a graph for GNN processing; determining one or more candidate node identifiers according to the root node identifier, wherein attribute data corresponding to the one or more candidate node identifiers are sequentially stored in a memory; and sampling one or more graph node identifiers at least from the one or more candidate node identifiers for the GNN processing.

Abstract: Methods, systems and apparatus, including computer programs encoded on computer storage medium, for controlling operations of machine tool workstations. Machine tool workstations are grouped into functional groups. Neural networks corresponding to the functional groups are trained to process respective inputs representing parts to be processed to generate respective outputs representing sequences of ordered subsets of the parts that produce a reduced setup time for workstations in the functional groups. Data representing respective collections of parts to be processed by workstations included in the functional groups is processed using the trained neural networks to generate corresponding sequences of ordered subsets of the collection of parts. Average delay times associated with the generated sequences of ordered subsets of the collection of parts are computed.

Abstract: The present disclosure discloses an image segmentation method combined with superpixel and multi-scale hierarchical feature recognition. This method is based on a convolutional neural network model taking multi-scale hierarchical features extracted from a Gaussian pyramid of an image as a recognition basis, and then being connected with a multilayer perceptron to achieve the recognition of each pixel in the image, moreover, this method is used tier performing superpixel segmentation on the image and is combined with a method for improving superpxiel in combination with LBP texture features to segment an original image so that an obtained superpixel block is more fitted to edges of targets, then, the original image is merged according to a mean value of a color, and finally, recognition of each target in the image is achieved.

Abstract: Methods, apparatus, and processor-readable storage media for automated resource prioritization using artificial intelligence techniques are provided herein.

Abstract: Methods, systems, apparatus and computer program products for implementing machine learning within control systems are disclosed. An industrial facility setting slate can be received from a machine learning system and a determination can be made as to whether to adopt the settings in the industrial facility setting slate. The machine learning model can be a neural network, e.g., a deep neural network, that has been trained, e.g., using reinforcement learning to predict a data setting slate that is predicted to optimize an efficiency of a data center.

Abstract: A method is disclosed for solving the Lagrangian dual of a constrained binary quadratic programming problem. The method comprises obtaining a constrained quadratic binary programming problem; until a convergence is detected, iteratively, performing a Lagrangian relaxation of the constrained quadratic binary programming problem to provide an unconstrained quadratic binary programming problem, providing the unconstrained quadratic binary programming problem to a quantum annealer, obtaining from the quantum annealer at least one corresponding solution, using the at least one corresponding solution to generate a new approximation for the Lagrangian dual bound; and providing a corresponding solution to the Lagrangian dual of the constrained binary quadratic programming problem after convergence.

Abstract: Information about a set of maintenance tasks and time windows includes a cost value per task per time window. Based on the information, a data model generator generates a data model, including task elements; time elements; cost elements; a total cost element; a constraint that requires each task element be assigned a time window from a respective domain, such that each time element is assigned a task count from a respective domain; a constraint that requires each cost element be assigned a cost value associated with a time window assigned (or to be assigned) to a task element corresponding to the cost element; and a constraint that requires the total cost element be assigned a total cost value that is a sum of the cost values assigned (or to be assigned) to the cost elements. Based on the data model, a CP solver determines a proposed maintenance schedule.

Abstract: A method may include receiving, for a package, shipment details including attributes, obtaining, for a subset of the attributes, logistic preferences, applying the logistic preferences to the shipment details to obtain modified shipment details, training a classifier using shipment transactions each including values for the attributes and labeled with a vendor logistic service, generating, by applying the classifier to the modified shipment details, scores for vendor logistic services, and recommending a vendor logistic service from the vendor logistic services using the scores.

Abstract: The invention discloses a MIMO different-factor compact-form model-free control method with parameter self-tuning. In view of the limitations of the existing MIMO compact-form model-free control method with the same-factor structure, namely, at time k, different control inputs in the control input vector can only use the same values of penalty factor and step-size factor, the invention proposes a MIMO compact-form model-free control method with the different-factor structure, namely, at time k, different control inputs in the control input vector can use different values of penalty factors and/or step-size factors, which can solve control problems of strongly nonlinear MIMO systems with different characteristics between control channels widely existing in complex plants. Meanwhile, parameter self-tuning is proposed to effectively address the problem of time-consuming and cost-consuming when tuning the penalty factors and/or step-size factors.

Abstract: Provided are a body part orientation estimation apparatus, a body part orientation estimation method, and a program that enable accurate body tracking without having the user wear many trackers. A time-series data input section (68) acquires a plurality of pieces of time-series data each representing positions, postures, or motions of a part of a body. The time-series data input section (68) inputs the plurality of pieces of time-series data into a conversion section (60). An output acquisition section (70) acquires a result of estimation of a position, a posture, or a motion of another part of the body that is closer to a center of the body than the part, the result of the estimation being an output obtained when the pieces of time-series data are input into the conversion section (60).

Abstract: An analysis system receives a time series. The data values of the time series correspond to a metric describing a characteristic of the computing system that changes over time. The analysis system stores a statistic value that represents the stationarity of the time series. In response to receiving a most recent value, the analysis system assigns the most recent value as the leading value in a window before retrieving the trailing value of the window. The analysis system updates the statistic value to add an influence of the most recent value and remove an influence of the trailing value. If the statistic value is less than a threshold, the analysis system determines that the time series is stationary. In response to determining the time series is stationary, the analysis system assigns an alert to the metric. The analysis system detects an anomaly in the metric based on the assigned alert.

Abstract: An example method of providing a configuration for a multitier microservice architecture includes receiving a configuration request from a user for a configuration that satisfies a set of conditions in a cloud environment. The method also includes searching a configuration data store for the configuration that matches the set of conditions. The configuration specifies a first container and a second container, the first container sends a first communication to the second container, and the second container sends a second communication responsive to the first communication to the first container. The method further includes in response to finding the configuration that matches the set of conditions: sending an allocation request to a cloud provider for allocation of the configuration in the cloud environment and providing a first identifier (ID) that identifies the first container and a second ID that identifies the second container to the user.

Abstract: An optimization device includes: a plurality of search parts; and a controller that controls the plurality of search parts, wherein, each of the plurality of search parts includes a state holding part configured to hold each of values of a plurality of state variables included in an evaluation function representing an energy value, an energy calculation part configured to calculate a change value of the energy value generated in a case where any one of the values of the plurality of state variables is changed, and a transition controller configured to stochastically determine whether or not to accept a state transition by a relative relation between the change value of the energy value and thermal excitation energy, based on a set temperature value, the change value, and a random number value.

Abstract: An apparatus to facilitate neural network (NN) training is disclosed. The apparatus includes training logic to receive one or more network constraints and train the NN by automatically determining a best network layout and parameters based on the network constraints.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating neural network architectures. One of the methods includes receiving a request to determine an architecture for a task neural network; maintaining data specifying a plurality of candidate architectures for the task neural network; repeatedly performing operations comprising: selecting one or more candidate architectures in the maintained data to be modified; generating a new candidate architecture from the selected candidate architecture by, for each hyperparameter in the set of hyperparameters, selecting the value for the hyperparameter for the new candidate architecture; and adding data specifying the new candidate architecture to the maintained data; and selecting, as the final architecture for the task neural network, one of the candidate architectures specified in the maintained data.

Abstract: A method is disclosed for solving the Lagrangian dual of a constrained binary quadratic programming problem. The method comprises obtaining a constrained quadratic binary programming problem; until a convergence is detected, iteratively, performing a Lagrangian relaxation of the constrained quadratic binary programming problem to provide an unconstrained quadratic binary programming problem, providing the unconstrained quadratic binary programming problem to a quantum annealer, obtaining from the quantum annealer at least one corresponding solution, using the at least one corresponding solution to generate a new approximation for the Lagrangian dual bound; and providing a corresponding solution to the Lagrangian dual of the constrained binary quadratic programming problem after convergence.

Abstract: An apparatus for training artificial intelligence (AI) models is presented. In embodiments, the apparatus may include an input interface to receive in real time model training data from one or more sources to train one or more artificial neural networks (ANNs) associated with the one or more sources, each of the one or more sources associated with at least one of the ANNs; a load distributor coupled to the input interface to distribute in real time the model training data for the one or more ANNs to one or more AI appliances; and a resource manager coupled to the load distributor to dynamically assign one or more computing resources on ones of the AI appliances to each of the ANNs in view of amounts of the training data received in real time from the one or more sources for their associated ANNs.

Abstract: The present invention relates to deep learning for automated segmentation of a medical image. More particularly, the present invention relates to deep learning for automated segmentation of anatomical regions and lesions in mammography screening and clinical assessment. According to a first aspect, there is provided a computer-aided method of segmenting regions in medical images, the method comprising the steps of: receiving input data; analysing the input data by identifying one or more regions; determining one or more characteristics for the one or more regions in the input data; and generating output segmentation data in dependence upon the characteristics for the one or more regions.

Abstract: Implementations of the present disclosure build a Bayesian student network using the knowledge learnt by an accurate but complex pre-trained teacher network, and sparsity induced by the variational parameters in a student network. Further, the sparsity inducing capability of the teacher on the student network is learnt by employing a Block Sparse Regularizer on a concatenated tensor of teacher and student network weights. Specifically, the student network is trained using the variational lower bound based loss function, constrained on the hint from the teacher, and block-sparsity of weights.

Abstract: In one embodiment, a model generator generates a new model for a behavior of a system based on an existing, authoritative model. First, a mapping generator generates a mapping model that maps authoritative values obtained via the authoritative model to measured values that represent the behavior of the system. Subsequently, the model generator creates the new model based on the authoritative model and the mapping model. In this fashion, the mapping model indirectly transforms the authoritative model to the new model based on the measured values. Advantageously, the authoritative model enables the model generator to increase a rate of accuracy improvement experienced while developing the new model compared to a rate of accuracy improvement that would be experienced were the new model to be generated based on conventional modeling techniques. In particular, for a given sampling budget, the model generator improves the accuracy of the new model.

Abstract: A method of managing memory usage of a stored training set for classification includes calculating one or both of a first similarity metric and a second similarity metric. The first similarity metric is associated with a new training sample and existing training samples of a same class as the new training sample. The second similarity metric is associated with the new training sample and existing training samples of a different class than the new training sample. The method also includes selectively storing the new training sample in memory based on the first similarity metric, and/or the second similarity metric.

Abstract: Systems and methods for optimizing factory scheduling, layout or both which represent active factory elements (human and machine) as computational objects and simulate factory operation to optimize a solution. This enables the efficient assembly of customized products, accommodates variable demand, and mitigates unplanned events (floor blockages, machines/IMRs/workcell/workers downtime, variable quantity, location, and destination of supply parts).

Abstract: The present invention discloses a prediction control method and system for component contents in a rare earth extraction process. The prediction control method includes: establishing an Elman neural network model of a rare earth extraction process; obtaining a predicted output value of the rare earth extraction process through the Elman neural network model of the rare earth extraction process; calculating an optimal set value through steady-state optimization; dynamically predicting an extractant flow increment and a detergent flow increment based on the predicted output value and the optimal set value; and controlling component contents in the rare earth extraction process according to the extractant flow increment and the detergent flow increment.

Abstract: A computer-implemented method according to one embodiment includes identifying a first classifier training data element and a second classifier training data element, calculating a similarity metric between the first classifier training data element and the second classifier training data element, and determining a classification for the first classifier training data element and the second classifier training data element, utilizing the similarity metric between the first classifier training data element and the second classifier training data element.

Abstract: In one embodiment, information (workload, performance, and configuration) is obtained about identified sub-systems (a target component plus other components that influence its performance). The identified sub-systems are clustered into workload clusters and also into performance clusters, where identified sub-systems of particular workload clusters have similar workload measurements, and identified sub-systems of particular performance clusters have similar performance metrics. The techniques herein then determine a given mapped performance cluster for a given workload cluster that corresponds to a best set of performance metrics from among all performance clusters mapped to the given workload cluster.

Abstract: A device can receive, from a user device, a request for a simulation of a forecasted change in production of a client organization. The device can obtain scheduling information for a set of historical work orders and a set of existing work orders associated with the client organization. The device can generate a set of simulated work orders using a forecasting technique, information included in the request, and the scheduling information. The device can generate one or more simulated schedules for simulating performance of the set of simulated work orders. The device can determine that one or more capacity values associated with operational resources needed to carry out the set of simulated schedules satisfy one or more threshold capacity values. The device can generate one or more recommendations for improving processes associated with the client organization. The device can provide the one or more recommendations to the user device.

Abstract: A method comprises a portable device obtaining a graphical encoded information item which is displayed on a display of a computing apparatus, decoding the encoded information from the encoded information item, and transmitting a first message to first server apparatus, the first message including the decoded information and a first identifier identifying the device or a user of the device, wherein the decoded information includes an apparatus identification information item for allowing identification of the computing apparatus, and the first server apparatus receiving the first message from the device, establishing the identity of the user of the device, wherein establishing the identity of the user comprises using the first identifier to determine if the user is registered with the first server apparatus in response to establishing the identity of the user, authorising the user to access a service, and providing the service to the user via the computing apparatus using the apparatus identification informati

Abstract: Disclosed herein are methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating an action selection policy for completing a task in an environment. The method includes identifying multiple possible actions in a state, wherein the state corresponds to a vector of information sets; identifying a vector of current action selection policies in the state, wherein each current action selection policy in the vector of current action selection policies corresponds to an information set in the vector of information sets; computing a sampling policy based on the vector of current action selection policies in the state; sampling an action among the multiple possible actions in the state according to a sampling probability of the action specified in the sampling policy; and updating each current action selection policy of the execution device in the state based on the action.

Abstract: An analytical method of determining the solution of an object with a plurality of planar surfaces under applied load includes the steps of assigning the load applied on each of the plurality of planar surfaces in the three dimensional directions respectively, determining the equivalent load applied on each of the plurality of planar surfaces in the three dimensional directions from the applied loads with the stress boundary conditions, and determining the stress and strain fields from the equivalent load applied on each of the plurality of planar surfaces.

Abstract: An equipment analysis support apparatus includes an equipment constraint information storage unit that stores equipment constraint information which is a correspondence relationship between each process constituting work performed using a plurality of pieces of equipment and a constraint to be imposed on the equipment in each process; a work process information acquisition unit that acquires work process information which is information on a process of the work currently in progress; a configuration information generation unit that, when it is determined that a current work state is changed, specifies a constraint to be imposed on the equipment in the changed work state and generates a configuration of analysis processing of the equipment satisfying the specified constraint based on the acquired work process information and the equipment constraint information; and a processing execution unit that performs processing necessary for analysis of the equipment based on the generated configuration.

Abstract: A method and system for determining a current context of a multimedia content element are provided. The method includes receiving at least one multimedia content element from a user device; receiving at least one environmental variable related to the at least one multimedia content element; generating at least one signature for the multimedia content element; determining a context of the at least one multimedia content element based on the at least one contextual parameter; and determining the current context of the at least one multimedia content element based on at least one contextual parameter and the determined context.

Abstract: A computer transforms high-dimensional data into low-dimensional data. A distance is computed between a selected observation vector and each observation vector of a plurality of observation vectors, a nearest neighbors are selected using the computed distances, and a first sigmoid function is applied to compute a distance similarity value between the selected observation vector and each of the selected nearest neighbors where each of the computed distance similarity values is added to a first matrix. The process is repeated with each observation vector of the plurality of observation vectors as the selected observation vector. An optimization method is executed with an initial matrix, the first matrix, and a gradient of a second sigmoid function that computes a second distance similarity value between the selected observation vector and each of the nearest neighbors to transform each observation vector of the plurality of observation vectors into the low-dimensional space.

Abstract: Generating a computing specification to be executed by a quantum processor includes: accepting a problem specification that corresponds to a second-quantized representation of a fermionic Hamiltonian, and transforming the fermionic Hamiltonian into a first qubit Hamiltonian including a first set of qubits that encode a fermionic state specified by occupancy of spin orbitals. An occupancy of any spin orbital is encoded in a number of qubits that is logarithmic in the number of spin orbitals, and a parity for a transition between any two spin orbitals is encoded in a number of qubits that is logarithmic in the number of spin orbitals. An eigenspectrum of a second qubit Hamiltonian, including the first set of qubits and a second set of qubit, includes a low-energy subspace and a high-energy subspace, and an eigenspectrum of the first qubit Hamiltonian is approximated by a set of low-energy eigenvalues of the low-energy subspace.

Abstract: Embodiments of the present disclosure disclose a method and apparatus for executing an instruction for an artificial intelligence chip. A specific embodiment of the method comprises: receiving descriptive information for describing a neural network model sent by a central processing unit, the descriptive information including at least one operation instruction; analyzing the descriptive information to acquire the at least one operation instruction; determining, for an operation instruction of the at least one operation instruction, a special-purpose execution component executing the operation instruction, and locking the determined special-purpose execution component; sending the operation instruction to the determined special-purpose execution component; and unlocking the determined special-purpose execution component in response to receiving a notification for instructing the operation instruction being completely executed.

Abstract: A wireless communications system includes a feedback processing unit for analyzing captured bandwidth data from a remote radio head, and a problem-type processor in operable communication with the feedback processing unit. The problem-type processor is configured to (i) analyze the captured bandwidth data to determine whether the captured bandwidth data presents one of a computational polynomial time problem and a non-deterministic polynomial-time hard (NP-hard) problem, and (ii) transmit problem-specific data based on the determination. The system further includes a communications processor in operable communication with the problem-type processor. The communications processor is configured to process polynomial time problem data from the transmitted problem-specific data. The system further includes a quantum computer in operable communication with the problem-type processor. The quantum computer is configured to process NP-hard problem data received from the transmitted problem-specific data.

Abstract: A system for arranging transport of adapted nutrimental artifacts with user-defined restriction requirements using artificial intelligence. The system includes at least a user-client device designed and configured to display at least an unrestricted nutrimental object, transmit at least a restricted nutrimental datum, transmit at least an adapted nutrimental request, and receive a selection of at least a sustenance provider and a selection of at least a physical performer. The system includes at least a server designed and configured to receive at least a restricted nutrimental datum. The system includes a nutrimental processing module operating on the at least a server designed and configured to generate at least a first filter set and transmit at least a first filter set.

Abstract: A method for dynamic intelligent scheduling includes following steps: collecting and recording resource constraints of multiple schedules on a production line and decision data of changes made to the schedules by a scheduler; cross-enumerating schedule combinations by using multiple production goals as penalty conditions; establishing a mathematical model based on the resource constraints and multi-objective weights corresponding to each schedule combination and importing the resource constraints to calculate schedule results; recording the penalty condition corresponding to the schedule combination matching the decision data as a valid penalty; using values of parameters corresponding to the valid penalty and values of the penalty conditions respectively as inputs and outputs to train a learning model; and responding to a scheduling request, finding a weight of each schedule combination by using the learning model according to the resource constraint of the current schedule and the production goals, and gene

Abstract: Methods are provided for determining discriminant functions of minimum risk linear classification systems, wherein a discriminant function is represented by a geometric locus of a principal eigenaxis of a linear decision boundary. A geometric locus of a principal eigenaxis is determined by solving a system of fundamental locus equations of binary classification, subject to geometric and statistical conditions for a minimum risk linear classification system in statistical equilibrium. Feature vectors and machine learning algorithms are used to determine discriminant functions and ensembles of discriminant functions of minimum risk linear classification systems, wherein distributions of the feature vectors have similar covariance matrices, and wherein a discriminant function of a minimum risk linear classification system exhibits the minimum probability of error for classifying given collections of feature vectors and unknown feature vectors related to the collections.

Abstract: Methods are provided for determining discriminant functions of minimum risk quadratic classification systems, wherein a discriminant function is represented by a geometric locus of a principal eigenaxis of a quadratic decision boundary. A geometric locus of a principal eigenaxis is determined by solving a system of fundamental locus equations of binary classification, subject to geometric and statistical conditions for a minimum risk quadratic classification system in statistical equilibrium. Feature vectors and machine learning algorithms are used to determine discriminant functions and ensembles of discriminant functions of minimum risk quadratic classification systems, wherein a discriminant function of a minimum risk quadratic classification system exhibits the minimum probability of error for classifying given collections of feature vectors and unknown feature vectors related to the collections.

Abstract: A method, computer program product, and computer system for transforming, by a computing device, a speech signal into a speech signal representation. A regression deep neural network may be trained with a cost function to minimize a mean squared error between actual values of the speech signal representation and estimated values of the speech signal representation, wherein the cost function may include one or more discriminative terms. Bandwidth of the speech signal may be extended by extending the speech signal representation of the speech signal using the regression deep neural network trained with the cost function that includes the one or more discriminative terms.

Abstract: A method of conducting computing experiments, includes executing a set of jobs, performing a comparison of a result of the executed set of jobs with templates of previously-executed experiments which are stored in a knowledge base, and identifying a prunable job of the set of jobs based on the comparison and a user constraint.

Abstract: Systems, methods, and non-transitory computer-readable media can identify a set of regions corresponding to a geographical area. A collection of training images can be acquired. Each training image in the collection can be associated with one or more respective recognized objects and with a respective region in the set of regions. Histogram metrics for a plurality of object categories within each region in the set of regions can be determined based at least in part on the collection of training images. A neural network can be developed based at least in part on the histogram metrics for the plurality of object categories within each region in the set of regions and on the collection of training images.

Abstract: An information processing apparatus including inter-class node insertion means for inserting an input vector into a network as an inter-class insertion node. The apparatus further includes a winner node learning time calculation means for incrementing, when an edge is connected between a first winner node and a second winner node, learning time of a node for the first winner node by a predetermined value.

Abstract: In various example embodiments, a comparative modeling system is configured to receive selections of a data set, a transform scheme, and one or more machine-learning algorithms. In response to a selection of the one or more machine-learning algorithms, the comparative modeling system determines parameters within the one or more machine-learning algorithms. The comparative modeling system generates a plurality of models for the one or more machine-learning algorithms, determines comparison metric values for the plurality of models, and causes presentation of the comparison metric values for the plurality of models.

Abstract: A neural network unit has at least one RAM, an output buffer and an array of neural processing units that: read first time step context layer node values from the output buffer; read second time step input layer node values from the RAM; generate second time step hidden layer node values based on the read input and context layer node values; output the hidden layer node values to the output buffer rather than to the RAM; read the hidden layer node values from the output buffer; generate second time step context layer node values based on the read hidden layer node values; output the context layer node values to the output buffer rather than to the RAM; generate output layer node values using the hidden layer node values; write the output layer node values to the RAM; and repeat for a sequence of time steps.

Abstract: A method for selecting a service endpoint from a plurality of service endpoints in a distributed system of a service provider may include storing processing data for each of the plurality of endpoints. A success rate may be calculated for each of the plurality of service endpoints and based on a number of processed requests from a plurality of received requests. An average latency may be calculated based on latency associated with each of the processed requests. A latency score may be calculated based on a minimum average latency and the average latency. A raw score may be calculated based on the latency score and the success rate. A selection weight may be calculated based on the raw score and a balancing parameter. One of the plurality of endpoints may be selected based on the selection weight.