-
公开(公告)号：US20240037391A1
公开(公告)日：2024-02-01
申请号：US18258523
申请日：2021-12-15
申请人： Orange
IPC分类号： G06N3/08
CPC分类号： G06N3/08
摘要： A method for distributed training of a graph-embedding neural network is disclosed. The method, performed at a first server, includes computing, based on a first input data sample, first model data and first embedding data of a first graph neural network, the first graph neural network corresponding to a first set of nodes of a graph that are visible to the first server, and includes sharing the first model data and the first embedding data with a second server. The method also includes receiving second embedding data from a third server, the second embedding data comprising embedding data of a second graph neural network corresponding to a second set of nodes of the graph that are invisible to the first server, and includes computing second model data of the first graph neural network based on a second input data sample and the embedding data of the second graph neural network.
-
2.发明公开公开(公告)号：US20240004707A1
公开(公告)日：2024-01-04
申请号：US18343304
申请日：2023-06-28
申请人： Orange
IPC分类号： G06F9/48
CPC分类号： G06F9/4893
摘要： An energy-efficient assignment of a task set T to a group of M processing devices models the process of deciding the assignment as a combinatorial optimization problem having an objective function optimizing the power consumption of the devices when executing subsets of the tasks, under a constraint that the total utilization of the task subset assigned to each respective processing device is lower than a threshold depending on the number of its processor cores. The objective function may be: min Σi=1M Pi(τi), where τi denotes the subset of tasks allocated to the ith device, and Pi(τi) represents power consumption of the ith device when executing τi, and the constraint Uτi≤Mi/4 for all the devices, where Uτi is the total utilization of τi executing on the ith device, and Mi denotes the number of cores of the ith device. Solving the problem using a MaxMin or genetic algorithm gives good energy efficiency.
-
3.发明公开公开(公告)号：US20240004706A1
公开(公告)日：2024-01-04
申请号：US18343219
申请日：2023-06-28
申请人： Orange
CPC分类号： G06F9/4893 , G06F9/5061
摘要： According to a directed-acyclic-graph representation, each real-time task is decomposed into sub-tasks, and a timing diagram is generated representing execution of the sub-tasks on a schedule respecting their deadlines and dependencies assuming an infinite number of processor cores. The timing diagram is segmented based on sub-task release times and deadlines, and each segment includes one or more parallel threads. For each segment, dependent on its workload, the frequency and/or voltage is selected to be used by the processor node(s) that are to execute the one or more threads of the segment, so as to reduce power consumption consistent with respecting the sub-task deadlines. Execution of the sub-tasks of the segments is then scheduled assuming the processor-core frequencies and/or voltages set in the deciding step, preferably by a global earliest deadline first algorithm.
-
-
搜索结果
3 条记录 (耗时 0.026 秒)
