CN113326917A - Method and system for automatically optimizing operator based on genetic algorithm - Google Patents
Method and system for automatically optimizing operator based on genetic algorithm Download PDFInfo
- Publication number
- CN113326917A CN113326917A CN202110475679.XA CN202110475679A CN113326917A CN 113326917 A CN113326917 A CN 113326917A CN 202110475679 A CN202110475679 A CN 202110475679A CN 113326917 A CN113326917 A CN 113326917A
- Authority
- CN
- China
- Prior art keywords
- scheduling
- individual
- operator
- fitness
- module
- 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
- 230000002068 genetic effect Effects 0.000 title claims abstract description 30
- 238000000034 method Methods 0.000 title claims abstract description 22
- 230000035772 mutation Effects 0.000 claims abstract description 23
- 238000012216 screening Methods 0.000 claims abstract description 22
- 238000003062 neural network model Methods 0.000 claims abstract description 14
- 238000004364 calculation method Methods 0.000 claims description 25
- 230000000903 blocking effect Effects 0.000 claims description 17
- 230000000717 retained effect Effects 0.000 claims description 7
- 238000005457 optimization Methods 0.000 description 13
- 238000013473 artificial intelligence Methods 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 150000004820 halides Chemical class 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 108090000623 proteins and genes Proteins 0.000 description 1
- 238000005215 recombination Methods 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Physiology (AREA)
- Genetics & Genomics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method and a system for automatically optimizing an operator based on a genetic algorithm, wherein a plurality of initial scheduling schemes of the operator are obtained, and each scheduling scheme is respectively used as an individual to form a population; respectively adopting each individual operation operator, and respectively calculating the fitness of each individual in the population according to the operation performance of the operator on the back-end equipment applying the neural network model; screening individuals in the population according to the fitness to obtain at least one reserved individual; judging whether a termination condition is reached: if yes, performing mutation operation on the reserved at least one individual to generate a new individual and form a new generation of population; and if not, outputting the individual with the highest fitness as a scheduling scheme of the operator, and then exiting. The scheduling strategy and the scheduling parameters thereof on the target hardware can be automatically searched out through a genetic algorithm, a high-performance operator is generated, and the development efficiency of the high-performance operator is greatly improved.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a method and a system for automatically optimizing an operator based on a genetic algorithm.
Background
Genetic algorithm is a heuristic search algorithm for solving the optimization, and the idea is to simulate the biological evolution process and to search for a better solution by selecting several locally optimal solutions and simulating some behaviors (cross-recombination, mutation) of the genetic genes. Genetic algorithms are widely applied in many scientific research and engineering fields at present because of being not constrained by restrictive assumptions of search space (function continuity is not required, and the properties of conductibility and the like) and the inherent implicit parallelism and global optimization capability.
At present, most artificial intelligence applications use deep learning technology. Whether the deep learning model can be successfully applied to the ground of the terminal meets the product requirements, and one key index is the reasoning performance of the neural network model. Good inference performance requires the deep learning inference framework to be implemented with high-performance optimization operators adapted to different hardware back-ends. However, most of the high-performance operators are optimized manually by advanced optimization engineers, and developing high-performance operators adapted to different backend is a process which takes a lot of manpower and time. Therefore, automation of high-performance operator development has become a relatively hot issue in recent years. Referring to fig. 1, the existing method is to separate the calculation implementation of the algorithm from the optimized scheduling policy, and the algorithm part is a pure algorithm calculation logic described by a Domain specific Language (Domain specific Language) and is independent of hardware. The optimized scheduling strategy part can cover the data loading delay by increasing the data parallelism and adjusting the calculation sequence, thereby obtaining the improvement on the performance. There are common scheduling strategies: vectorization, namely vectorization calculation is carried out on a certain dimension, and each n elements are used as a vector to be calculated; parallelization, namely performing thread-level parallelization on a certain dimension; and (4) block calculation, namely dividing a certain cycle into two cycles, namely an inner cycle and an outer cycle. Although the prior art improves the development efficiency of optimization operators to a certain extent, and directly specifies an optimization strategy to generate codes at a corresponding back end, the optimization scheduling strategy needs to be manually selected, the scheduling parameters of the optimization scheduling strategy are determined, and different parameters are needed for different hardware.
Disclosure of Invention
The invention provides a method and a system for automatically optimizing an operator based on a genetic algorithm, and aims to solve the technical problems of low efficiency of manual optimization scheduling strategies and parameters and the like in the prior art.
A method for automatically optimizing an operator based on a genetic algorithm is applied to a neural network model, the neural network model comprises an operator for executing the genetic algorithm, and the method comprises the following steps:
step S1, acquiring a plurality of initial scheduling schemes of operators, and taking each scheduling scheme as an individual to form a population;
step S2, each individual operation operator is adopted, and the fitness of each individual in the population is calculated according to the operation performance of the operator on the back-end equipment applying the neural network model;
s3, screening individuals in the population according to the fitness to obtain at least one reserved individual;
step S4, determining whether the termination condition is reached:
if yes, go to step S6;
if not, go to step S5;
step S5, performing mutation operation on the retained at least one individual to generate new individuals and compose a new generation of population, and then returning to step S2;
and step S6, outputting the individuals with the highest fitness as a scheduling scheme of the operator, and then exiting.
Further, the scheduling scheme includes a scheduling policy of the operator and a scheduling parameter adapted to the scheduling policy.
Further, in step S3, selecting the individuals with the highest fitness to complete the population screening;
then in step S5, mutation operation is performed on the individuals with the highest fitness remaining to generate new individuals.
Further, in step S3, eliminating the individuals with the lowest fitness to complete the screening of the population;
then in step S5, mutation operations are performed on all individuals that are retained to generate new individuals, respectively.
Further, in step S4, the termination condition is:
whether the current iteration times reach the preset maximum iteration times or not; or whether the current highest fitness reaches the preset qualified fitness.
Further, in step S5, one of the following mutation operations is randomly selected for mutation to obtain a new individual;
when the scheduling strategy comprises the blocking operation, the sizes of an inner loop and an outer loop formed after the blocking operation are exchanged;
changing the execution sequence among a plurality of scheduling strategies executed in sequence in the scheduling strategies; and
and changing the parameter scale of the scheduling parameter corresponding to the scheduling strategy.
A system for automatically optimizing an operator based on a genetic algorithm, the system comprising:
the system comprises a setting module, a dispatching module and a dispatching module, wherein the setting module is used for a plurality of initial dispatching schemes of operators, and each dispatching scheme is respectively used as an individual to form a population;
the fitness calculation module is connected with the setting module and is used for respectively adopting each individual operation operator and respectively calculating the fitness of each individual according to the operation performance of the operator on the back-end equipment applying the neural network model;
the screening module is connected with the fitness calculation module and used for screening the individuals in the population according to the fitness to obtain at least one reserved individual;
the judging module is connected with the screening module and used for judging whether the terminal condition of the fitness calculation of the individuals in the population is reached or not and outputting a judging result;
the variation module is connected with the judgment module and used for performing variation operation on at least one reserved individual to generate a new individual and form a new generation of population when the judgment result is that the termination condition is reached;
the fitness calculation module is also used for connecting the variation module and calculating the fitness of each individual in the new population;
and the output module is connected with the variation module, is used for being connected with the judgment module and is used for outputting the individual with the highest fitness.
Further, the scheduling scheme includes a scheduling policy of the operator and a scheduling parameter adapted to the scheduling policy.
Further, the termination conditions are:
whether the current iteration times reach the preset maximum iteration times or not; or whether the current highest fitness reaches the preset qualified fitness.
Further, the variation module is used for randomly selecting one of the following variation operations to perform variation to obtain a new individual;
when the scheduling strategy comprises the blocking operation, the sizes of an inner loop and an outer loop formed after the blocking operation are exchanged;
changing the execution sequence among a plurality of scheduling strategies executed in sequence in the scheduling strategies; and
and changing the parameter scale of the scheduling parameter corresponding to the scheduling strategy.
The beneficial technical effects of the invention are as follows: compared with the traditional calculation, the method and the system for automatically optimizing the operator based on the genetic algorithm save the manual designated scheduling strategy, can automatically search the scheduling strategy and the scheduling parameters thereof on the target hardware through the genetic algorithm to generate the high-performance operator, and greatly improve the development efficiency of the high-performance operator.
Drawings
FIG. 1 is a block diagram of a prior art optimization operator;
FIG. 2 is a flow chart illustrating the steps of the method for automatically optimizing an operator based on a genetic algorithm according to the present invention;
FIG. 3 is a block diagram of the method for automatically optimizing an operator based on a genetic algorithm according to the present invention;
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the embodiments and features of the embodiments may be combined with each other without conflict.
The invention is further described with reference to the following drawings and specific examples, which are not intended to be limiting.
Referring to fig. 2, a method for automatically optimizing an operator based on a genetic algorithm, applied to a neural network model, wherein the neural network model includes an operator for executing the genetic algorithm, is characterized by comprising the following steps:
step S1, acquiring a plurality of initial scheduling schemes of operators, and taking each scheduling scheme as an individual to form a population;
step S2, each individual operation operator is adopted, and the fitness of each individual in the population is calculated according to the operation performance of the operator on the back-end equipment applying the neural network model;
s3, screening individuals in the population according to the fitness to obtain at least one reserved individual;
step S4, determining whether the termination condition is reached:
if yes, go to step S6;
if not, go to step S5;
step S5, performing mutation operation on the retained at least one individual to generate new individuals and compose a new generation of population, and then returning to step S2;
and step S6, outputting the individuals with the highest fitness as a scheduling scheme of the operator, and then exiting.
Specifically, the fitness refers to the inverse of the running time of the operator at the back end of the device, that is, the shorter the time is, the higher the performance is, and the higher the fitness is.
Further, the scheduling scheme includes a scheduling policy of the operator and a scheduling parameter adapted to the scheduling policy.
The scheduling strategies of the operators comprise scheduling strategies in the aspects of high dimensionality, wide dimensionality and channel number dimensionality, and the scheduling strategies of the high dimensionality, the wide dimensionality and the channel number dimensionality respectively have corresponding scheduling parameters.
Specifically, the scheduling strategy comprises vectorization calculation, parallelization calculation and loop expansion calculation; the scheduling parameters include a vectorization scale, a parallelization scale, and a loop unrolling scale.
Further, in step S3, selecting the individuals with the highest fitness to complete the population screening;
then in step S5, mutation operation is performed on the individuals with the highest fitness remaining to generate new individuals.
Specifically, the screening of the individuals in the population includes ranking the fitness from large to small to form a rank of each individual, and screening the individual with the maximum fitness.
Further, in step S3, eliminating the individuals with the lowest fitness to complete the screening of the population;
then in step S5, mutation operations are performed on all individuals that are retained to generate new individuals, respectively.
And screening the individuals in the population to eliminate the individuals with the lowest fitness.
Further, in step S4, the termination condition is:
whether the current iteration times reach the preset maximum iteration times or not; or whether the current highest fitness reaches the preset qualified fitness.
Further, in step S5, one of the following mutation operations is randomly selected for mutation to obtain a new individual;
when the scheduling strategy comprises the blocking operation, the sizes of an inner loop and an outer loop formed after the blocking operation are exchanged;
changing the execution sequence among a plurality of scheduling strategies executed in sequence in the scheduling strategies; and
and changing the parameter scale of the scheduling parameter corresponding to the scheduling strategy.
And changing the execution sequence among the scheduling strategies which are executed in sequence into the sequence of exchanging parallelization calculation and vectorization calculation.
Referring to fig. 3, a system for automatically optimizing an operator based on a genetic algorithm, applied to the system for automatically optimizing an operator based on a genetic algorithm, includes:
the system comprises a setting module (1) and a control module, wherein the setting module is used for a plurality of initial scheduling schemes of operators, and each scheduling scheme is used as an individual to form a population;
the fitness calculation module (2) is connected with the setting module (1) and is used for respectively adopting each individual operation operator and respectively calculating the fitness of each individual according to the operation performance of the operator on the back-end equipment applying the neural network model;
the screening module (3) is connected with the fitness calculation module (2) and is used for screening the individuals in the population according to the fitness to obtain at least one reserved individual;
the judging module (4) is connected with the screening module (3) and used for judging whether the terminal condition of fitness calculation of individuals in the population is reached or not and outputting a judging result;
the variation module (5) is connected with the judgment module (4) and is used for performing variation operation on at least one reserved individual to generate a new individual and form a new population when the judgment result is that the termination condition is reached;
the fitness calculation module (2) is also used for connecting the variation module (5) and calculating the fitness of each individual in the new population;
and the result output module (6) is connected with the variation module (5) and is used for being connected with the judgment module (4) and outputting the individual with the highest fitness.
Further, the scheduling scheme includes a scheduling policy of the operator and a scheduling parameter adapted to the scheduling policy.
Further, the termination conditions are:
whether the current iteration times reach the preset maximum iteration times or not; or whether the current highest fitness reaches the preset qualified fitness.
Further, the mutation module (5) is used for randomly selecting one of the following mutation operations to carry out mutation to obtain a new individual;
when the scheduling strategy comprises the blocking operation, the sizes of an inner loop and an outer loop formed after the blocking operation are exchanged;
changing the execution sequence among a plurality of scheduling strategies executed in sequence in the scheduling strategies; and
and changing the parameter scale of the scheduling parameter corresponding to the scheduling strategy.
Specifically, the optimization operator is a convolution operator, and an individual scheduling policy of the convolution operator may be defined as: vectorizing a high h dimension, parallelizing a channel number c dimension, and circularly expanding a wide w dimension, preferably, the vectorization scale is 4, the parallelization scale is 8, and the circularly expanding scale is 4.
Specifically, AutoKernel is used as a high-performance operator automatic optimization tool.
Specifically, in the Halide scheduling policy, loop unrolling (unoll) is an optimization method for increasing the execution speed of a program by sacrificing the size of the program, and aims to: the register is more fully utilized, the times of loading and saving each operation memory during circulation are reduced, the utilization rate of the register is improved, the register is fully utilized to cache data, the delay of accessing the memory can be greatly reduced, and the memory bandwidth is effectively improved.
Specifically, vectorization (vectore) in the Halide scheduling strategy is to convert several scalar calculations (scales) into one vector calculation (vector), and fully utilize SIMD vector instructions. Most modern CPUs support SIMD (Single Instruction Multiple Data) operation. SIMD can perform the same operation/instruction (e.g., add, multiply, etc.) on multiple values simultaneously in the same CPU cycle.
Specifically, the chunk (tile) in the hide scheduling policy refers to a loop chunk (loop blocking), and a large loop is changed into two loops: the outer loop and the inner loop of the small blocks are used for mining the time and space locality of the innermost loop, so that the performance is improved.
Specifically, the parameter scale of the scheduling parameter corresponding to the scheduling policy is changed, for example, the cyclic expansion scale of the width w dimension of one individual is changed to 4, and after the mutation operation, the cyclic expansion scale is changed to 8.
Specifically, the sizes of the inner loop and the outer loop formed after the blocking operation are exchanged, for example, one individual blocking operation is to block the H-dimension, the H-dimension is 128, the blocking size is 16, the outer loop (H _ outer ═ 128/tile _ size ═ 8) after the blocking, the inner loop (H _ inner ═ tile _ size ═ 16), the inner loop and the outer loop after the mutation operation are of the sizes, the outer loop (H _ outer ═ 16), and the inner loop (H _ inner ═ 8).
While the invention has been described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the spirit and scope of the invention.
Claims (10)
1. A method for automatically optimizing an operator based on a genetic algorithm is applied to a neural network model, the neural network model comprises an operator for executing the genetic algorithm, and the method is characterized by comprising the following steps:
step S1, obtaining a plurality of initial scheduling schemes of the operator, and taking each scheduling scheme as an individual to form a population;
step S2, respectively adopting each individual to operate the operator, and respectively calculating the fitness of each individual in the population according to the operation performance of the operator on the back-end equipment applying the neural network model;
step S3, screening the individuals in the population according to the fitness to obtain at least one reserved individual;
step S4, determining whether the termination condition is reached:
if yes, go to step S6;
if not, go to step S5;
a step S5 of performing mutation operation on at least one of the individuals retained to generate new individuals and compose a new generation of the population, and then returning to the step S2;
and step S6, outputting the individual with the highest fitness as the scheduling scheme of the operator, and then exiting.
2. The method of claim 1, wherein the scheduling scheme comprises a scheduling policy of the operator and scheduling parameters adapted to the scheduling policy.
3. The method for automatically optimizing an operator based on a genetic algorithm according to claim 1, wherein in step S3, the individuals with the highest fitness are selected to complete the screening of the population;
then in step S5, mutation operation is performed on the individuals with the highest fitness that are retained to generate new individuals.
4. The method for automatically optimizing operators based on genetic algorithms according to claim 1, wherein in the step S3, the individuals with the lowest fitness are eliminated to complete the screening of the population;
then in step S5, mutation operations are performed on all the retained individuals to generate new individuals.
5. The method for automatically optimizing an operator based on a genetic algorithm according to claim 1, wherein in step S4, the termination condition is:
whether the current iteration times reach the preset maximum iteration times or not; or whether the current highest fitness reaches the preset qualified fitness.
6. The method for automatically optimizing an operator based on a genetic algorithm as claimed in claim 2, wherein in said step S5, one of the following mutation operations is randomly selected to mutate to obtain a new individual;
when the scheduling strategy comprises the blocking operation, the sizes of an inner loop and an outer loop formed after the blocking operation are exchanged;
changing an execution order among a plurality of scheduling policies executed in order among the scheduling policies; and
and changing the parameter scale of the scheduling parameter corresponding to the scheduling strategy.
7. A system for automatically optimizing an operator based on a genetic algorithm, wherein the system for automatically optimizing an operator based on a genetic algorithm according to any one of claims 1 to 6 is applied, and comprises:
the system comprises a setting module, a scheduling module and a judging module, wherein the setting module is used for a plurality of initial scheduling schemes of operators, and each scheduling scheme is respectively used as an individual to form a population;
the fitness calculation module is connected with the setting module and is used for respectively adopting each individual to operate the operator and respectively calculating the fitness of each individual according to the operation performance of the operator on the back-end equipment applying the neural network model;
the screening module is connected with the fitness calculation module and used for screening the individuals in the population according to the fitness to obtain at least one reserved individual;
the judging module is connected with the screening module and used for judging whether the terminal condition of the fitness calculation of the individuals in the population is reached or not and outputting a judging result;
the variation module is connected with the judgment module and is used for performing variation operation on at least one reserved individual to generate a new individual and form a new generation of the population when the judgment result is that the termination condition is reached;
the fitness calculation module is further used for connecting the variation module and calculating the fitness of each individual in the new population;
and the output module is connected with the variation module, is used for being connected with the judgment module and is used for outputting the individual with the highest fitness.
8. The system for automatically optimizing an operator based on a genetic algorithm as claimed in claim 7, wherein the scheduling scheme comprises a scheduling policy of the operator and scheduling parameters adapted to the scheduling policy.
9. The system for automatically optimizing an operator based on a genetic algorithm according to claim 7, wherein the termination condition is:
whether the current iteration times reach the preset maximum iteration times or not; or whether the current highest fitness reaches the preset qualified fitness.
10. The system according to claim 8, wherein said mutation module is configured to randomly select one of the mutation operations to mutate to obtain a new individual;
when the scheduling strategy comprises the blocking operation, the sizes of an inner loop and an outer loop formed after the blocking operation are exchanged;
changing an execution order among a plurality of scheduling policies executed in order among the scheduling policies; and
and changing the parameter scale of the scheduling parameter corresponding to the scheduling strategy.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110475679.XA CN113326917A (en) | 2021-04-29 | 2021-04-29 | Method and system for automatically optimizing operator based on genetic algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110475679.XA CN113326917A (en) | 2021-04-29 | 2021-04-29 | Method and system for automatically optimizing operator based on genetic algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113326917A true CN113326917A (en) | 2021-08-31 |
Family
ID=77413959
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110475679.XA Pending CN113326917A (en) | 2021-04-29 | 2021-04-29 | Method and system for automatically optimizing operator based on genetic algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113326917A (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH09251446A (en) * | 1996-03-18 | 1997-09-22 | Nri & Ncc Co Ltd | Optimizing device and method thereof by genetic algorithm based on unbalanced evolution theory |
| CN106960244A (en) * | 2017-03-14 | 2017-07-18 | 北京无线电测量研究所 | A kind of genetic algorithm and the evolution algorithm of particle cluster algorithm Parallel Fusion |
| CN107392497A (en) * | 2017-08-08 | 2017-11-24 | 大连交通大学 | A kind of Job-Shop system and dispatching method based on improved adaptive GA-IAGA |
| WO2018161468A1 (en) * | 2017-03-10 | 2018-09-13 | 东莞理工学院 | Global optimization, searching and machine learning method based on lamarck acquired genetic principle |
| CN109800071A (en) * | 2019-01-03 | 2019-05-24 | 华南理工大学 | A kind of cloud computing method for scheduling task based on improved adaptive GA-IAGA |
| CN110389819A (en) * | 2019-06-24 | 2019-10-29 | 华中科技大学 | A kind of dispatching method and system of computation-intensive batch processing task |
| CN111767977A (en) * | 2020-06-09 | 2020-10-13 | 中国人民解放军国防科技大学 | Group particle gradient descent algorithm based on improved genetic algorithm |
| CN112181598A (en) * | 2020-10-12 | 2021-01-05 | 内蒙古农业大学 | Cloud computing task scheduling method based on improved genetic algorithm |
-
2021
- 2021-04-29 CN CN202110475679.XA patent/CN113326917A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH09251446A (en) * | 1996-03-18 | 1997-09-22 | Nri & Ncc Co Ltd | Optimizing device and method thereof by genetic algorithm based on unbalanced evolution theory |
| WO2018161468A1 (en) * | 2017-03-10 | 2018-09-13 | 东莞理工学院 | Global optimization, searching and machine learning method based on lamarck acquired genetic principle |
| CN106960244A (en) * | 2017-03-14 | 2017-07-18 | 北京无线电测量研究所 | A kind of genetic algorithm and the evolution algorithm of particle cluster algorithm Parallel Fusion |
| CN107392497A (en) * | 2017-08-08 | 2017-11-24 | 大连交通大学 | A kind of Job-Shop system and dispatching method based on improved adaptive GA-IAGA |
| CN109800071A (en) * | 2019-01-03 | 2019-05-24 | 华南理工大学 | A kind of cloud computing method for scheduling task based on improved adaptive GA-IAGA |
| CN110389819A (en) * | 2019-06-24 | 2019-10-29 | 华中科技大学 | A kind of dispatching method and system of computation-intensive batch processing task |
| CN111767977A (en) * | 2020-06-09 | 2020-10-13 | 中国人民解放军国防科技大学 | Group particle gradient descent algorithm based on improved genetic algorithm |
| CN112181598A (en) * | 2020-10-12 | 2021-01-05 | 内蒙古农业大学 | Cloud computing task scheduling method based on improved genetic algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10963787B2 (en) | Systems and methods for generation of sparse code for convolutional neural networks | |
| CN105975342A (en) | Improved cuckoo search algorithm based cloud computing task scheduling method and system | |
| CN107357652A (en) | A kind of cloud computing method for scheduling task based on segmentation sequence and standard deviation Dynamic gene | |
| CN114741955A (en) | Multi-objective optimization task scheduling method based on security cloud | |
| CN110362780B (en) | Large data tensor canonical decomposition calculation method based on Shenwei many-core processor | |
| Ji et al. | Parallelizing word2vec in multi-core and many-core architectures | |
| Subramoney et al. | Multi-swarm PSO algorithm for static workflow scheduling in cloud-fog environments | |
| Wahib et al. | Optimization of parallel genetic algorithms for nVidia GPUs | |
| CN112434785B (en) | Distributed parallel deep neural network performance evaluation method for supercomputer | |
| CN106293003A (en) | A kind of heterogeneous system dynamic power consumption optimization method based on AOV gateway key path query | |
| CN115016938A (en) | Calculation graph automatic partitioning method based on reinforcement learning | |
| Xiao et al. | FCNNLib: An efficient and flexible convolution algorithm library on FPGAs | |
| CN117271101B (en) | Operator fusion method and device, electronic equipment and storage medium | |
| Pérez et al. | Parallel/distributed implementation of cellular training for generative adversarial neural networks | |
| CN112990461B (en) | Method, device, computer equipment and storage medium for constructing neural network model | |
| Zhao et al. | GPU implementation of a cellular genetic algorithm for scheduling dependent tasks of physical system simulation programs | |
| CN113326917A (en) | Method and system for automatically optimizing operator based on genetic algorithm | |
| CN116842992A (en) | Operator fusion method and system for image recognition depth neural network | |
| WO2021054990A1 (en) | Systems and methods for generation of sparse code for convolutional neural networks | |
| CN114356738A (en) | Method for predicting time required for executing neural network model and related product | |
| KR20230012790A (en) | Method and apparatus for function optimization | |
| Jiang et al. | A genetic algorithm for scheduling in heterogeneous multicore system integrated with FPGA | |
| Akbari et al. | An efficient algorithm for compile-time task scheduling problem on heterogeneous computing systems | |
| Kornelsen et al. | Fast heterogeneous task mapping for reducing edge dnn latency | |
| CN111274030B (en) | Efficient multiprocessor system-on-chip design space mining method oriented to application features |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |