EP3931696A1 - Method and system for using computing resources of a multiprocessor computing system - Google Patents
Method and system for using computing resources of a multiprocessor computing systemInfo
- Publication number
- EP3931696A1 EP3931696A1 EP20707268.7A EP20707268A EP3931696A1 EP 3931696 A1 EP3931696 A1 EP 3931696A1 EP 20707268 A EP20707268 A EP 20707268A EP 3931696 A1 EP3931696 A1 EP 3931696A1
- Authority
- EP
- European Patent Office
- Prior art keywords
- software
- execution
- priority
- tasks
- software tasks
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 48
- 238000013401 experimental design Methods 0.000 claims abstract description 20
- 238000013439 planning Methods 0.000 claims abstract description 10
- 238000004364 calculation method Methods 0.000 claims description 42
- 238000002474 experimental method Methods 0.000 claims description 36
- 238000013461 design Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 11
- 238000004458 analytical method Methods 0.000 claims description 8
- 238000004422 calculation algorithm Methods 0.000 claims description 5
- 238000005457 optimization Methods 0.000 description 9
- 238000004088 simulation Methods 0.000 description 8
- 238000007726 management method Methods 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000004590 computer program Methods 0.000 description 3
- 230000003247 decreasing effect Effects 0.000 description 3
- 239000000463 material Substances 0.000 description 3
- 238000010200 validation analysis Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 2
- 239000004575 stone Substances 0.000 description 2
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 238000004026 adhesive bonding Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000012938 design process Methods 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 238000013383 initial experiment Methods 0.000 description 1
- 230000007786 learning performance Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 238000012913 prioritisation Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5061—Partitioning or combining of resources
- G06F9/5066—Algorithms for mapping a plurality of inter-dependent sub-tasks onto a plurality of physical CPUs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
- G06F9/4806—Task transfer initiation or dispatching
- G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
- G06F9/4881—Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/461—Saving or restoring of program or task context
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
- G06F9/4806—Task transfer initiation or dispatching
- G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
- G06F9/485—Task life-cycle, e.g. stopping, restarting, resuming execution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5011—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resources being hardware resources other than CPUs, Servers and Terminals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2209/00—Indexing scheme relating to G06F9/00
- G06F2209/48—Indexing scheme relating to G06F9/48
- G06F2209/483—Multiproc
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2209/00—Indexing scheme relating to G06F9/00
- G06F2209/50—Indexing scheme relating to G06F9/50
- G06F2209/503—Resource availability
Definitions
- TITLE Method and system for using computing resources of a multiprocessor computing system
- the present invention relates to a method and a system for using computing resources of a computing system comprising a plurality of interconnected microprocessors and adapted to operate in parallel, to perform software tasks.
- the invention lies in the field of optimizing computing resources in microprocessor "clusters", and finds particular application in the use of such computing resources to solve complex physical problems, for example to simulate physical phenomena, by using iterative calculations implementing the resolution of mathematical equations modeling physical phenomena.
- the invention is applicable in the field of computer simulation, in particular in the context of simulation methods which consume a lot of resources, or in the computational resolution of problems in fluid dynamics.
- the invention proposes a method of using the computing resources of a computing system comprising a plurality of interconnected microprocessors and adapted to operate in parallel, to perform software tasks. This process uses:
- an experiment plan comprising a plurality of software tasks to be performed in order to solve a predetermined physical problem, defined by at least one input parameter and at least one output parameter, the experiment plan being calculated as a function of a predetermined initial computational budget, said software tasks of the experimental plan having a first level of priority,
- the method makes it possible to achieve an optimized occupation of the available computation resources.
- the method of using computing resources according to the invention may exhibit one or more of the characteristics below, taken independently or in any acceptable combination:
- Obtaining freed computation resources comprises an analysis of available computational resources, and in the event of no available computational resource, an analysis of the first priority level software tasks currently being executed, an execution stop of at least part of said first level software tasks and a backup of an associated execution context.
- obtaining freed up compute resources includes a license availability check available to perform a second priority software task.
- the method further comprises, in the event of the absence of at least one software task of second priority level awaiting execution, a resumption of execution of software tasks of first priority level placed in execution stop.
- the distribution of at least part of the first priority level software tasks over the available computing resources comprises an analysis of available computing resources, comprising an analysis of available computation queues, and a distribution of the first priority level software tasks on selected calculation queues.
- the process of calculating an experiment design involves a supervised learning algorithm of a statistical meta-model.
- the statistical meta-model is a regression model by Gaussian process, and the process of calculating an experimental design involves an entropy maximization of a kernel of covariance of said Gaussian process.
- the calculation of an experimental design comprises a determination of an experimental space from the input parameters of the problem to be solved, a mesh of said experimental space in calculation points, and a determination of a set of calculation points according to an associated computational budget.
- the calculation of an experiment plan is iterated based on the execution results obtained from an execution of a previous experiment plan, and in which at each iteration, a current computational budget is calculated.
- the current computational budget is equal, at each iteration, to half of a remaining computational budget calculated based on the initial computational budget and a computational budget consumed during previous iterations.
- the iteration of calculating an experiment design to solve a predetermined physical problem is stopped when a stop criterion is verified, in particular when the remaining computational budget is consumed.
- the invention relates to the use of the resources of a computing system comprising a plurality of interconnected microprocessors and adapted to operate in parallel, to perform software tasks.
- This system includes:
- a module for calculating an experiment plan comprising a plurality of software tasks to be carried out to solve a physical problem defined by at least one input parameter and at least one output parameter, the experiment plan being calculated by as a function of a predetermined initial computational budget, said software tasks to be performed having a first level of priority,
- a module for planning the execution of software tasks by the computing system configured for:
- Figure 1 schematically illustrates a system for using the computing resources of a microprocessor farm according to one embodiment
- FIG. 2 is a block diagram of the processes implemented in one embodiment of a method for using the computing resources according to one embodiment
- Figure 3 is a block diagram of the steps for generating experimental designs according to one embodiment.
- the invention will be described below in its application to the simulation and calculations associated with solving physical problems, performed as sequences of software tasks, but is not limited to this field of application.
- the computing resource utilization system 1 of Figure 1 comprises a farm (or "cluster") of interconnected microprocessors forming a computing system 2.
- microprocessors are divided into interconnected computation nodes and grouped together in different computation queues, each having an associated computational capacity.
- the cluster 2 consists of a plurality of interconnected programmable electronic devices, for example computers, each programmable electronic device comprising an electronic memory unit and a central computing unit, or CPU, comprising one or more electronic microprocessors, adapted to communicate via a communication bus.
- each programmable electronic device is produced in the form of programmable logic components, such as an FPGA (standing for Field-Programmable Gâte Arra ⁇ ), or else in the form of dedicated integrated circuits, of the ASIC type (for Application-Specific Integrated Circuit).
- Programmable electronic devices are suitable for executing software tasks, from computer code instructions in executable format, in any appropriate format.
- a software task is understood here to mean the execution of a computer program on a set of input data.
- System 1 also includes a module 4 for planning the execution of software tasks.
- the module 4 is produced in the form of an executable computer program, stored in a memory unit 6 of a programmable electronic device 8 forming part of the cluster 2, and executable by a processor of the programmable electronic device 8.
- the memory unit 6 also stores a module 10 for formatting the results of execution of software tasks.
- the module 4 for scheduling the execution of the software tasks of the system 1 is configured to manage the execution of the software tasks 20 of the first level of priority, also called non-priority tasks, and of the software tasks 22 of the second level of priority, also. called priority tasks.
- priority software tasks come from user commands.
- the non-priority software tasks are associated with physical problems to be studied Ri. .Rg, and are intended to be executed when space is available on the cluster.
- Each physical problem to be studied P L is defined by a set of data, which are for example stored in a file 24 stored on a physical medium readable by computer, and supplied at the input of the system 1.
- a physical problem is defined by:
- a parameter noted ParameterJ is between the minimum limits Val_min_i and maximum Val_max_i;
- the resolution of the physical problem consists in studying the values of a parameter or of several parameters of output, considered to be parameters of interest, as a function of the values of the input parameters.
- Non-exhaustive examples of such physical phenomena are, for example: calculation of the time required to completely melt a metal paving stone subjected to a heat source on a wall, as a function of the heating power, of the outside temperature, of the heat capacity of the material and dimensions of the paving stone; based on an angle and size of a deflector, calculate an associated pressure loss; as a function of the input parameters: speed of entry into the computation domain, dynamic viscosity, geometric dimension of the computation domain, calculate a pressure variation at the limits of the computation domain, gluing length.
- the computational budget is, in one embodiment, a maximum number of software tasks to be executed for the resolution of the physical problem.
- the computational budget also includes the minimum number of microprocessors to be taken into account for the computation.
- the calculation code is scientific calculation software, looking for a solution to equations representing physical phenomena iteratively.
- such software is subject to a runtime license, and the number of parallel runs is limited to the number of runtime licenses available.
- Each PL physical problem defined by a set of data as described above, is processed by a module 30 for calculating the design of the experiment.
- the module 30 is produced in the form of a computer program that can be executed by one of the programmable electronic devices of the cluster 2.
- Module 30 implements an experiment design process for each physical problem to be addressed.
- An experiment plan is defined by a set of simulations to be carried out, each simulation implementing a plurality of software tasks to be executed, with a set of parameters chosen as input, each simulation making it possible to obtain a value for each output parameter chosen forming an execution result 32.
- the execution result is provided, for example in raw form, for example a binary file, to an operator, via a suitable interface of the system, for example as a computer file.
- the execution result 32 in raw form is transmitted to the module 10 for formatting the execution results of software tasks, which generates formatted execution results 34.
- These formatted results contain the input parameters and the parameters. output of the problem, which are extracted, if any, from the raw execution results, which include a lot of data.
- the experiment design calculation module 30 is adapted to iteratively calculate successive experiment designs for the resolution of a given physical problem, taking into consideration, for the calculation of a current experiment design, formatted results 34 resulting from the execution of the software tasks implemented for the preceding steps of the experimental design.
- FIG. 2 is a block diagram of the processes implemented by the module 4 for planning the execution of software tasks.
- Module 4 implements a step 40 for obtaining the state of the hardware resources, in particular the availability of computing resources in the interconnected microprocessors of cluster 2.
- step 40 the number of microprocessors available among the plurality of interconnected microprocessors is obtained in step 40. This step is repeated at regular time intervals, for example every minute.
- the module 4 also implements a step 42 of receiving requests for execution of first priority software tasks (non-priority software tasks) submitted by the module 30 for calculating the design of the experiment, and a step 44 for checking the presence second priority software task execution requests (priority software tasks). For example, non-priority software tasks are submitted in a first computation queue, and priority software tasks are submitted in a second computation queue.
- Steps 42 and 44 are for example carried out substantially in parallel, and repeated at regular time intervals.
- module 4 retrieves (step 46) information relating to the current availability of computing resources.
- the availability of computing resources here includes the availability of computing microprocessors and also the availability of software licenses, for example by a license token mechanism, to perform priority software tasks.
- step 54 the execution of the priority task or tasks is started (step 54).
- step 46 is followed by a step 48 of freeing computational resources to perform the priority software tasks.
- Step 48 comprises a sub-step 50 for analyzing the non-priority software tasks being executed, and for selecting non-priority software tasks to stop, as a function of the computing resources and the licenses consumed.
- the software tasks of a plan of experiment implemented for the resolution of a given physical problem are selected.
- any license token released by stopping execution of one of the non-priority tasks is made available.
- the module 4 for planning the execution of the software tasks detects the presence of priority software tasks awaiting execution, it obtains the computational resources freed to allow the execution of these software tasks in step d execution 54.
- the priority software task or tasks are distributed over the calculation queues comprising available calculation resources.
- module 4 checks (step 56) for the presence of computing resources, including license resources, available.
- non-priority software tasks that were stopped previously are resumed (step 58), if applicable.
- the non-priority software tasks awaiting execution, forming part of experiment plans submitted by the experiment plan calculation module 30, are loaded and executed on the available computing resources (step 60).
- a calculation queue is selected for the execution of a non-priority software task, according to the state of the resources of the cluster, for example according to the number of available microprocessors. and the number of calculations to be performed for the execution of the software task (s).
- Non-priority software tasks in a given experiment plan are executed until a stop criterion is validated.
- the validation of the stop criterion corresponds to the achievement of the predetermined computational budget in terms of the number of computations or to the achievement of a target learning performance value of the problem.
- the non-priority software tasks are determined by the module 30 for calculating experimental designs, an operation of which is described below with reference to FIG. 3.
- the occupation of the computing resources is optimized, while at the same time. maintaining the prioritization of certain priority software tasks, in particular software tasks whose processing is required by a user.
- FIG. 3 is a block diagram of the processes implemented by the experiment design calculation module 30.
- a first step 70 the data defining the physical problem to be solved are obtained.
- this data is provided by an operator.
- this is the only intervention of a human operator, the generation of the successive experimental plans being performed automatically by the module 30.
- a PL physical problem is defined by:
- a parameter noted ParameterJ is between the minimum limits Val_min_i and maximum Val_max_i;
- the computational budget is a maximum number of software tasks to be performed to solve the PL problem.
- An experimental space is defined by the number N L of input parameters, which is a positive integer, depending on the problem PL to be solved.
- the experimental space is a hyper-pad of dimension N L , possibly reduced by the user according to the unnecessary combinations of parameters.
- a mesh of this experimental space in calculation points is then defined.
- the mesh is isotropic.
- the mesh is defined so that the total number of calculation points is less than a predetermined limit value, for example equal to 10 5 .
- step 72 additional information is provided during a step 72, for example information making it possible to refine the mesh of the experimental space.
- the initial experimental design to solve the physical problem PL is calculated, based on the remaining computational budget.
- the remaining computational budget is the value corresponding to the initial computational budget minus the computational budget already consumed. On initialization, the remaining computational budget is equal to the initial computational budget.
- a computational budget for the experimental design is current is computed.
- the remaining computational budget is then also equal to B L / 2.
- an initial experiment plan is designed in step 76, consisting in choosing a number of calculation points NP L, o on which the calculation code is to be executed, forming a set of software tasks to be executed, according to of the available computational budget ML , O.
- the design of the experiment plan is based on an optimization procedure described in detail below with reference to steps 84 and 86, focused on an optimality criterion of the experiment plan, for example a maximum entropy criterion (in the sense of Shannon's information theory). Of course, this is an example of an optimality criterion, other criteria could be used.
- the experimental design is initialized by random drawing in the experiment space and then enhanced by the optimization procedure.
- the experimental design is built point by point with the objective of optimizing the same criteria as the optimization procedure for each addition.
- the experiment plan thus generated is transmitted to module 4 for planning the execution of software tasks, and executed by cluster 2.
- Formatted execution results 34 are received in step 78 following execution of the experimental design.
- execution results are processed by the module 30 to update knowledge of the PL problem to be treated, in order to allow optimization of the generation of experiment plans.
- the influence of the input parameters on the values of the output parameters, known from the run results, is analyzed, so as to refine the selection of calculation points to generate the experimental design.
- the module 30 constructs a mathematical meta-model and implements a supervised statistical learning algorithm as a function of the execution results.
- the statistical meta-model is a Gaussian process regression model.
- Gaussian process regression models are well known in the field of statistical data processing. They assimilate the covariance of the Gaussian random variable representative of the modeled process to an analytical function, also called kernel, parameterized by scalar values called hyperparameters.
- the module 30 optimizes the hyperparameters of the covariance kernel, for example by the maximum likelihood method.
- the regression by Gaussian process, provided with an a priori kernel of covariance kg is linked to the optimality criterion of the experimental design.
- the optimality criterion consists in maximizing the determinant of the matrix K such that:
- Ki j kg (x ir Xj ) where i, X j are calculation points of the experimental design.
- the validation of the stop criterion is verified during a step 80.
- the stop criterion is validated when the initial computational budget has been consumed, or, in other words, when the remaining computational budget is at 0. In this case, step 80 is followed by stopping the calculations (step 82).
- reaching a performance target is also a stop criterion.
- reaching a performance target is also a stop criterion.
- the performance criterion is the average value of the coefficient of determination of the linear regression between the exact result of the simulation on the one hand, and the result predicted by the meta-model on the other hand, the latter being trained on a selection of other cases, for example by the "10-fold cross-validation" method.
- the target value of the performance criterion is then set when the user submits the problem, and can typically be of the order of 0.95.
- the computational budget for the current iteration i c is calculated, and in one embodiment, it is equal to half of the remaining computational budget.
- computation points are selected (step 84), and a current experimental design is calculated (step 86).
- an exchange procedure optimization algorithm configured to increase the entropy of Shannon information is implemented.
- Exchange candidate calculation points are selected as follows.
- a number PL, ÎC of “endangered” calculation points is determined, for example a percentage of the N L , i calculation points of the current experimental plan. For example, this percentage is between 10% and 30%.
- the normalized covariance matrix K is calculated, and the inverse matrix K -1 called the precision matrix is calculated.
- PL, ÎC calculation points are selected. These points of calculation are the points having the least mutual information with the other points of calculation of the experimental plan previously executed.
- the experiment design calculation module 30 implements supervised statistical methods for an automated resolution of the problems to be solved, while controlling the calculation budget and without human intervention.
- the process implemented by the calculation module makes it possible to perform calculations with control over material resources and execution time thanks to the implementation of the calculation budget.
- the module 4 for planning the execution of the software tasks implements an optimization of the use of resources, while managing the priorities of the software tasks to be executed.
- the system 1 for the use of computational resources including these two collaborating modules allows optimized and controlled management of costs, including material cost, license cost and cost in human intervention, computation times and resource optimization being optimized.
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Stored Programmes (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Multi Processors (AREA)
Abstract
Description
Claims
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
FR1902132A FR3093366B1 (en) | 2019-03-01 | 2019-03-01 | Method and system for using computing resources of a multiprocessor computing system |
PCT/EP2020/055266 WO2020178168A1 (en) | 2019-03-01 | 2020-02-28 | Method and system for using computing resources of a multiprocessor computing system |
Publications (1)
Publication Number | Publication Date |
---|---|
EP3931696A1 true EP3931696A1 (en) | 2022-01-05 |
Family
ID=67660203
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
EP20707268.7A Pending EP3931696A1 (en) | 2019-03-01 | 2020-02-28 | Method and system for using computing resources of a multiprocessor computing system |
Country Status (5)
Country | Link |
---|---|
US (1) | US20220147387A1 (en) |
EP (1) | EP3931696A1 (en) |
CA (1) | CA3131756A1 (en) |
FR (1) | FR3093366B1 (en) |
WO (1) | WO2020178168A1 (en) |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8607242B2 (en) * | 2010-09-02 | 2013-12-10 | International Business Machines Corporation | Selecting cloud service providers to perform data processing jobs based on a plan for a cloud pipeline including processing stages |
US9513962B2 (en) * | 2013-12-03 | 2016-12-06 | International Business Machines Corporation | Migrating a running, preempted workload in a grid computing system |
US10691488B2 (en) * | 2017-12-01 | 2020-06-23 | International Business Machines Corporation | Allocating jobs to virtual machines in a computing environment |
US11379263B2 (en) * | 2018-08-13 | 2022-07-05 | Ares Technologies, Inc. | Systems, devices, and methods for selecting a distributed framework |
-
2019
- 2019-03-01 FR FR1902132A patent/FR3093366B1/en active Active
-
2020
- 2020-02-28 CA CA3131756A patent/CA3131756A1/en active Pending
- 2020-02-28 WO PCT/EP2020/055266 patent/WO2020178168A1/en active Application Filing
- 2020-02-28 US US17/435,519 patent/US20220147387A1/en active Pending
- 2020-02-28 EP EP20707268.7A patent/EP3931696A1/en active Pending
Also Published As
Publication number | Publication date |
---|---|
FR3093366B1 (en) | 2021-03-12 |
US20220147387A1 (en) | 2022-05-12 |
CA3131756A1 (en) | 2020-09-10 |
FR3093366A1 (en) | 2020-09-04 |
WO2020178168A1 (en) | 2020-09-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11875262B2 (en) | Learning neural network structure | |
US20200401891A1 (en) | Methods and apparatus for hardware-aware machine learning model training | |
US20220188677A1 (en) | Systems and methods implementing an intelligent machine learning tuning system providing multiple tuned hyperparameter solutions | |
US20200050443A1 (en) | Optimization and update system for deep learning models | |
EP3286647A1 (en) | Placement of a calculation task on a functionally asymmetric processor | |
CN104035747B (en) | Method and apparatus for parallel computation | |
FR2982386A1 (en) | METHOD, COMPUTER PROGRAM, AND CLUSTER COMPUTER RESOURCE ALLOCATION DEVICE FOR EXECUTING A WORK SUBMITTED TO AUDIT CLUSTER | |
CN104679590A (en) | Map optimization method and device in distributive calculating system | |
CN113112013A (en) | Optimized quantization for reduced resolution neural networks | |
US20210174202A1 (en) | Method and apparatus with model optimization, and accelerator system | |
Monge et al. | Ensemble learning of runtime prediction models for gene-expression analysis workflows | |
Yu et al. | Heterogeneous federated learning using dynamic model pruning and adaptive gradient | |
WO2020178168A1 (en) | Method and system for using computing resources of a multiprocessor computing system | |
EP3588301A1 (en) | Automatic and auto-optimised determination of parameters for executing a software application on an information processing platform | |
EP3506201A1 (en) | Adaptive system and method for automatically tracking at least one target in at least one video stream | |
WO2023185842A1 (en) | Method for determining compile optimization option, electronic device and readable storage medium | |
EP3506202B1 (en) | Adaptive system and method for automatically tracking at least one target in at least one video stream | |
CN115398455A (en) | Neural network system for distributed promotion of programmable logic controller by using multiple processing units | |
CN114202033A (en) | Coal classification method, system and storage medium | |
CN113128682A (en) | Automatic neural network model adaptation method and device | |
US20220335285A1 (en) | Methods, apparatus, and articles of manufacture to improve performance of an artificial intelligence based model on datasets having different distributions | |
FR3122759A1 (en) | IMPLEMENTATIONS AND METHODS OF PROCESSING NEURONAL NETWORK IN SEMICONDUCTOR HARDWARE | |
US20230153394A1 (en) | One-pass approach to automated timeseries forecasting | |
US20230297848A1 (en) | Optimizing cascade of classifiers schema using genetic search | |
EP3923208A1 (en) | Automated hardware resource optimization |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: UNKNOWN |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE |
|
PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |
|
17P | Request for examination filed |
Effective date: 20210831 |
|
AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |
|
DAV | Request for validation of the european patent (deleted) | ||
DAX | Request for extension of the european patent (deleted) | ||
P01 | Opt-out of the competence of the unified patent court (upc) registered |
Effective date: 20230606 |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: EXAMINATION IS IN PROGRESS |
|
17Q | First examination report despatched |
Effective date: 20240226 |