Abstract: A method for requesting a recommendation for content items in a recommender system is provided. The method includes extracting a user attribute from the recommender system, identifying multiple content items that match a user preference based on the user attribute, determining an interest value for the user on each content item based on an affinity between the user attribute and a content item attribute, and providing the user a list of content items ranked according to the interest value for the user. A system including a memory storing instructions and a processor configured to execute the instructions and cause the system to perform the above method is also provided.

Abstract: Disclosed methods, systems, and apparatus, include computer programs encoded on computer storage media, for performing allocation of M resources among N users into K pools by solving a knapsack problem (KP) using a distributed computing system. The method includes: receiving data representing K global constraints and L local constraints of the KP; transforming the KP into a dual problem using K dual multipliers; decomposing the dual problem into N sub-problems; performing two or more iterations in solving the dual problem, wherein in one iteration, for each dual multiplier corresponding to a global constraint corresponding to a pool: determining an updated dual multiplier for the global constraint corresponding to the pool to be a non-negative threshold; and computing M decision variables of each of the N users corresponding to the updated dual multiplier in solving each of the N sub-problems corresponding to the each of the N users.

Abstract: Disclosed methods, systems, and apparatus, include computer programs encoded on computer storage media, for performing allocation of M resources among N users into K pools by solving a knapsack problem (KP) using a distributed computing system. The method includes: receiving data representing K global constraints and L local constraints of the KP; transforming the KP into a dual problem using K dual multipliers; decomposing the dual problem into N sub-problems; performing two or more iterations in solving the dual problem, wherein in one iteration, for each dual multiplier corresponding to a global constraint corresponding to a pool: determining an updated dual multiplier for the global constraint corresponding to the pool to be a non-negative threshold; and computing M decision variables of each of the N users corresponding to the updated dual multiplier in solving each of the N sub-problems corresponding to the each of the N users.

Abstract: An example system includes prediction workers, training workers, and a parameter server. The prediction workers store a local copy of a machine-learned model and run the mode exclusively in serving mode. The training workers store a local copy of a machine-learned model and a local snapshot and run the local copy exclusively in training mode and compare the local model or state to the snapshot after training to send delta updates to the parameter server after training. The parameter server aggregates received delta updates into a master copy of the model, sends the aggregated updates back to training workers and provides two types of updates; a real-time update based on a comparison of the master model with a local snapshot, and a full update. The real-time update occurs at least an order of magnitude more frequently than the full update and includes a subset of the weights in the model.

Abstract: Disclosed methods, systems, and apparatus, include computer programs encoded on computer storage media, for performing allocation of M resources among N users into K pools by solving a knapsack problem (KP) using a distributed computing system. The method includes: receiving data representing K global constraints and L local constraints of the KP; transforming the KP into a dual problem using K dual multipliers; decomposing the dual problem into N sub-problems; performing two or more iterations in solving the dual problem, wherein in one iteration, for each dual multiplier corresponding to a global constraint corresponding to a pool: determining an updated dual multiplier for the global constraint corresponding to the pool to be a non-negative threshold; and computing M decision variables of each of the N users corresponding to the updated dual multiplier in solving each of the N sub-problems corresponding to the each of the N users.

Abstract: Disclosed methods, systems, and apparatus, include computer programs encoded on computer storage media, for performing allocation of M resources among N users into K pools by solving a knapsack problem (KP) using a distributed computing system. The method includes: receiving data representing K global constraints and L local constraints of the KP; transforming the KP into a dual problem using K dual multipliers; decomposing the dual problem into N sub-problems; performing two or more iterations in solving the dual problem, wherein in one iteration, for each dual multiplier corresponding to a global constraint corresponding to a pool: determining an updated dual multiplier for the global constraint corresponding to the pool to be a non-negative threshold; and computing M decision variables of each of the N users corresponding to the updated dual multiplier in solving each of the N sub-problems corresponding to the each of the N users.

Abstract: Disclosed herein are methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing allocation of M resources among N users into K pools by solving a knapsack problem (KP) using a distributed computing system that includes a number of individual solvers. The method includes: receiving data representing K global constraints and L local constraints of the KP; decomposing the KP into N sub-problems using K dual multipliers, each of the N sub-problems corresponding to a respective one of the N users and subject to the L local constraints w.r.t. the corresponding user, wherein N is in an order of billions or larger; determining the number of individual solvers for solving the N sub-problems; distributing the N sub-problems among the number of individual solvers; and solving the KP by the distributed computing system by performing two or more iterations.

Abstract: Disclosed methods, systems, and apparatus, include computer programs encoded on computer storage media, for performing allocation of M resources among N users into K pools by solving a knapsack problem (KP) using a distributed computing system. The method includes: receiving data representing K global constraints and L local constraints of the KP; transforming the KP into a dual problem using K dual multipliers; decomposing the dual problem into N sub-problems; performing two or more iterations in solving the dual problem, wherein in one iteration, for each dual multiplier corresponding to a global constraint corresponding to a pool: determining an updated dual multiplier for the global constraint corresponding to the pool to be a non-negative threshold; and computing M decision variables of each of the N users corresponding to the updated dual multiplier in solving each of the N sub-problems corresponding to the each of the N users.

Abstract: Disclosed herein are methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing allocation of M resources among N users into K pools by solving a knapsack problem (KP) using a distributed computing system that includes a number of individual solvers. The method includes: receiving data representing K global constraints and L local constraints of the KP; decomposing the KP into N sub-problems using K dual multipliers, each of the N sub-problems corresponding to a respective one of the N users and subject to the L local constraints w.r.t. the corresponding user, wherein N is in an order of billions or larger; determining the number of individual solvers for solving the N sub-problems; distributing the N sub-problems among the number of individual solvers; and solving the KP by the distributed computing system by performing two or more iterations.

Abstract: Disclosed methods, systems, and apparatus, include computer programs encoded on computer storage media, for performing allocation of M resources among N users into K pools by solving a knapsack problem (KP) using a distributed computing system. The method includes: receiving data representing K global constraints and L local constraints of the KP; transforming the KP into a dual problem using K dual multipliers; decomposing the dual problem into N sub-problems; performing two or more iterations in solving the dual problem, wherein in one iteration, for each dual multiplier corresponding to a global constraint corresponding to a pool: determining an updated dual multiplier for the global constraint corresponding to the pool to be a non-negative threshold; and computing M decision variables of each of the N users corresponding to the updated dual multiplier in solving each of the N sub-problems corresponding to the each of the N users.

Abstract: Disclosed herein are methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing allocation of M resources among N users into K pools by solving a knapsack problem (KP) using a distributed computing system that includes a number of individual solvers. The method includes: receiving data representing K global constraints and L local constraints of the KP; decomposing the KP into N sub-problems using K dual multipliers, each of the N sub-problems corresponding to a respective one of the N users and subject to the L local constraints w.r.t. the corresponding user, wherein N is in an order of billions or larger; determining the number of individual solvers for solving the N sub-problems; distributing the N sub-problems among the number of individual solvers; and solving the KP by the distributed computing system by performing two or more iterations.

Abstract: Method, system, and programs for recommending content to a user. First information related to one or more previous users is first obtained. A model that maps from users to topics of interest is then established based on the first information related to the one or more previous users. Second information related to the current user is also obtained. One or more topics of interest are identified for the current user based on the model. Content is recommended to the current user in accordance with the one or more topics of interest for the current user. Eventually, an updated model is generated by integrating information associated with the current user with the model established based on the first information related to the one or more previous users. The information associated with the current user includes the second information related to the current user.

Abstract: Disclosed herein are methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing allocating M resources subject to L constraints. The method includes: receiving data representing L constraints, wherein each of the L constraints corresponds to a subset of M resources and restricts a respective maximum number C of resources to be selected among the subset of the M resources, wherein the L constraints has a hierarchal structure; determining a topological ordering of the L constraints; selecting all the M resources as an initial selection; removing resources from the initial selection by traversing each constraint in the topological ordering of the L constraints; and allocating the selected resources after traversing all the L constraints in the topological ordering of the L constraints.

Abstract: Method, system, and programs for recommending content to a user. First information related to one or more previous users is first obtained. A model that maps from users to topics of interest is then established based on the first information related to the one or more previous users. Second information related to the current user is also obtained. One or more topics of interest are identified for the current user based on the model. Content is recommended to the current user in accordance with the one or more topics of interest for the current user. Eventually, an updated model is generated by integrating information associated with the current user with the model established based on the first information related to the one or more previous users. The information associated with the current user includes the second information related to the current user.