Abstract: Techniques for generating a set of Deep Learning (DL) models are described. An example method includes training an initial set of DL models using the training data, wherein a topology of each of the DL models is determined based on the parameters vector. The method also includes generating a set of estimate performance functions for each of the DL models in the initial set based on the set of edge-related metrics, and generating a plurality of objective functions based on the set of estimated performance functions. The method also includes generating a final DL model set based on the objective functions, receiving a user selection of a selected DL model from the final DL model set, and deploying the selected DL model to an edge device.

Abstract: A method, apparatus and a product for generating a dataset for a reinforcement model. The method comprises obtaining a plurality of different subsets of the set of features; for each subset of features, determining a policy using a Markov Decision Process; obtaining a state comprises a valuation of each feature of the set of features; applying the plurality of policies on the state, whereby obtaining a plurality of suggested actions for the state, based on different projections of the state onto different subsets of features; determining, for the state, one or more actions and corresponding scores thereof based on the plurality of suggested actions; and training a reinforcement learning model using the state and the one or more actions and corresponding scores thereof.

Abstract: Constructing a MARS prediction model using predictor variables at a first point in time within a time horizon, including directly-controllable variables of first physical characteristics of a system and that are associated with adjustable operational control settings for directly controlling the first physical characteristic, and including controllable variables of second physical characteristics that are affected by the first physical characteristics, recursively using the prediction model to define an optimization problem for later point in time within the time horizon, transforming the optimization problem into a MILP problem, and solving the MILP problem using an optimization engine to determine, for any given one of the directly-controllable variables and corresponding to at least one of the points in time, for adjusting, using the optimized value, the adjustable operational control setting corresponding to the given directly-controllable variable and thereby control the physical characteristic associate

Abstract: In an approach, a processor stores a dictionary set, including simplex dictionaries, for saving processing time when calculating an optimal control policy for at least one linearly time dependent value function of a plurality of variables complying with a plurality of linearly time dependent constraints. A processor calculates a storage limit for the dictionary set, based on a number of the plurality of variables, the plurality of constraints, and size of a memory. A processor removes at least one of the plurality of simplex dictionaries in accordance with the storage limit, while maintaining a neighbor density measure, where the neighbor density measure is based on a distance between the at least one of the simplex dictionaries and a non-removed simplex dictionary and the distance corresponds to a number of simplex pivots required to construct the at least one of the simplex dictionaries from the non-removed simplex dictionary.

Abstract: A method for predicting at least one score for at least one item, comprising in at least one iteration of a plurality of iterations: receiving a user profile having a plurality of user attribute values; computing the at least one score according to a similarity between the user profile and a plurality of other user profiles by inputting the user profile and a plurality of items into a prediction model trained by: in each of a plurality of training iterations: receiving a training user profile of a plurality of training user profiles, the training user profile having a plurality of training user attribute values; computing by the prediction model a plurality of predicted scores, each for one of a plurality of training items, in response to the training user profile and the plurality of training items, where each of the plurality of training items has a plurality of training item.

Abstract: A method, apparatus and a product for generating a dataset for a reinforcement model. The method comprises obtaining a plurality of different subsets of the set of features; for each subset of features, determining a policy using a Markov Decision Process; obtaining a state comprises a valuation of each feature of the set of features; applying the plurality of policies on the state, whereby obtaining a plurality of suggested actions for the state, based on different projections of the state onto different subsets of features; determining, for the state, one or more actions and corresponding scores thereof based on the plurality of suggested actions; and training a reinforcement learning model using the state and the one or more actions and corresponding scores thereof.

Abstract: Automatic identification of features that drive a reinforcement learning model to recommend an action of interest. The identification is based on a calculation of occupation measures of state-action pairs associated with the reinforcement learning model. High occupation measures of certain action-state pairs indicate that the states of these pairs likely include the sought-after features.

Abstract: A heterogeneous resource reservation (HRR) manager configured to classify historical application requests from a past time interval for a first workload to generate labeled historical application requests. The HRR manager further configured to generate a forecast based on the labeled historical application requests and for predicting future application requests for the first workload for a future time interval and calculate a joint plan based on the forecast. The joint plan including a set of virtual resources, a set of billing contracts, and a set of load balancer weights. The HRR manager further configured to implement the joint plan for a distributed computing workload during the future time interval.

Abstract: Techniques for generating a set of Deep Learning (DL) models are described. An example method includes training an initial set of DL models using the training data, wherein a topology of each of the DL models is determined based on the parameters vector. The method also includes generating a set of estimate performance functions for each of the DL models in the initial set based on the set of edge-related metrics, and generating a plurality of objective functions based on the set of estimated performance functions. The method also includes generating a final DL model set based on the objective functions, receiving a user selection of a selected DL model from the final DL model set, and deploying the selected DL model to an edge device.

Abstract: Constructing a MARS prediction model using predictor variables at a first point in time within a time horizon, including directly-controllable variables of first physical characteristics of a system and that are associated with adjustable operational control settings for directly controlling the first physical characteristic, and including controllable variables of second physical characteristics that are affected by the first physical characteristics, recursively using the prediction model to define an optimization problem for later point in time within the time horizon, transforming the optimization problem into a MILP problem, and solving the MILP problem using an optimization engine to determine, for any given one of the directly-controllable variables and corresponding to at least one of the points in time, for adjusting, using the optimized value, the adjustable operational control setting corresponding to the given directly-controllable variable and thereby control the physical characteristic associate

Abstract: A heterogeneous resource reservation (HRR) manager configured to classify historical application requests from a past time interval for a first workload to generate labeled historical application requests. The HRR manager further configured to generate a forecast based on the labeled historical application requests and for predicting future application requests for the first workload for a future time interval and calculate a joint plan based on the forecast. The joint plan including a set of virtual resources, a set of billing contracts, and a set of load balancer weights. The HRR manager further configured to implement the joint plan for a distributed computing workload during the future time interval.

Abstract: An example method includes receiving a nominal equivalent resource usage data, an infrastructure usage data, an effective production capacity, a demand elasticity curve, and workload scheduling constraints across a plurality of accounts. The method includes calculating an equivalent resource utilization based on the nominal equivalent resource usage data, the infrastructure usage data, and the effective production capacity. The method includes calculating a potential value increase for a service based on the workload scheduling constraints, the nominal equivalent resource usage data, the effective production capacity, and the demand elasticity curve. The method includes calculating a value increase scheme for the service based on the potential value increase and sending the value increase scheme to a user workload device.

Abstract: A method comprising obtaining a scheduling problem comprising: a set of agents and a set of tasks to be performed by the set of agents, wherein solving the scheduling problem using an automated solver is not feasible using available predetermined resources. The method comprises automatically generating a plurality of alternative scheduling problems, wherein a solution to each such alternative scheduling problem defines a solution to the scheduling problem and determining a solution to the scheduling problem by applying the automated solver to solve, while using the available predetermined resources, an alternative problem of the plurality of alternative scheduling problems to determine a solution to the alternative problem and by mapping the solution to the alternative problem to the scheduling problem, whereby determining the solution.

Abstract: A method comprising using at least one hardware processor for receiving sensory data from at least one physical or virtual sensor. The hardware processor(s) are used for computing a plurality of decision options for configuration of the at least one physical or virtual sensor. The hardware processor(s) are used for computing a plurality of utility functions, and for each utility function: (a) computing a utility value for each decision option, and (b) identifying a first subset of decision options that substantially maximize the computed utility values. The hardware processor(s) are used for selecting at least one cross-function decision option from of the first subsets, wherein the at least one cross-function decision option is included in a substantially maximum number of the first subsets. The hardware processor(s) are used for applying at least one of the at least one cross-function decision options, to at least one physical or virtual sensor.

Abstract: Method, system and product for decomposing a simulation model. The method comprising automatically decomposing the simulation model into a predetermined number of co-simulation components, wherein each co-simulation component is allocated to a different simulation platform, wherein said automatically decomposing comprises: defining a target optimization function, wherein the target optimization function computes an estimated run time of the simulation model, wherein the target optimization function is based on a communication time within each co-simulation component and a communication time between each pair of co-simulation components; and determining a decomposition of the simulation model that optimizes a value of the target optimization function. The method further comprises executing the decomposed simulation model by executing in parallel each co-simulation component on a different simulation platform, whereby the simulation model is executed in a distributed manner.

Abstract: An example method includes receiving a nominal equivalent resource usage data, an infrastructure usage data, an effective production capacity, a demand elasticity curve, and workload scheduling constraints across a plurality of accounts. The method includes calculating an equivalent resource utilization based on the nominal equivalent resource usage data, the infrastructure usage data, and the effective production capacity. The method includes calculating a potential value increase for a service based on the workload scheduling constraints, the nominal equivalent resource usage data, the effective production capacity, and the demand elasticity curve. The method includes calculating a value increase scheme for the service based on the potential value increase and sending the value increase scheme to a user workload device.

Abstract: A method, system, and product for simulation of Internet of Things (IoT) environment. The method performed by a simulation node in the IoT environment, which comprises the simulation node and a cloud server connected by a computerized network. The method comprises selecting a simulated IoT device to simulate from a plurality of simulated IoT devices that are being simulated by the simulation node; invoking a real-world model to obtain real-world simulated values; determining a simulated behavior of the selected simulated IoT device by invoking a device model and providing the real-world simulated values thereto, o wherein the simulated behavior comprises transmitting a message to the cloud server; setting a next simulated action of the simulation node to occur at a designated time, wherein the next simulated action is the simulated behavior; and performing the next simulated action at the designated time.

Abstract: A computer program product comprising a non-transitory computer readable storage medium retaining program instructions configured to cause a processor to perform actions, which program instructions implement: a framework for creating a model of an NP-hard problem, the model comprising at least one entity selected from the group comprising: an objective, a variable, an equation and a constraint, wherein the framework provides methods for automatically transforming the model, comprising: one or more methods for manipulating or changing a status of the entity of the model, the methods comprising a method for imposing or ignoring the constraint; and one or more methods related to operations to be applied to the entity of the model.

Abstract: A computer program product, an apparatus and a method for training of an HMM. The method comprises applying a classifier that uses an HMM which was trained based on a training set, on a set of samples to provide an initial prediction; computing a first F1-score of the initial prediction measuring an accuracy of the initial prediction; selecting a misclassified sample by the classifier in the initial prediction; adding the misclassified sample to the training set; training the HMM using the misclassified sample to provide a modified HMM; applying the classifier using the modified HMM on the set of samples to provide a second prediction; computing a second F1-score of the second prediction; and comparing the first F1-score and the second F1-score; in response to a determination that the first F1-score is greater than the second F1-score, removing the misclassified sample from the training set.

Abstract: A computer-implemented method, computerized apparatus and computer program product for selecting time windows to vehicle routing problems. A set of criteria for estimating desirability of scheduling an appointment to a time interval, and a set of time intervals at which appointments can be scheduled are obtained. A new appointment for scheduling to a time interval is received. For each time interval of the set, a balanced score according to the set of criteria is calculated. A time interval for scheduling the new appointment is selected based on the balanced score.