Acceleration system and dynamic configuration method thereof
An acceleration system includes a plurality of modules. Each of the plurality of modules includes at least one central processing unit, at least one graphics processing unit, at least one field programmable gate array, or at least one application specific integrated circuit. At least one of the plurality of modules includes at least another of the plurality of modules such that the acceleration system is structural and nested.
1. An acceleration system, comprising:
a plurality of modules, wherein each of the plurality of modules comprises at least one central processing unit, at least one graphics processing unit, at least one field programmable gate array, or at least one application specific integrated circuit,
wherein at least one of the plurality of modules comprises at least another of the plurality of modules, such that the acceleration system is structured and nested,
wherein numbers, hierarchies, or types of the plurality of modules of the acceleration system is dynamically configured according to a recurrent structural causal model.
2. The acceleration system of claim 1 , wherein an input data is divided into a plurality of groups, the plurality of groups correspond to the plurality of modules of the acceleration system respectively, and the plurality of modules of the acceleration system process the plurality of groups corresponding to the plurality of modules respectively.
3. The acceleration system of claim 2 , wherein the input data is divided into the plurality of groups according to a clustering algorithm.
4. The acceleration system of claim 2 , wherein a first module within the plurality of modules comprises a second module within the plurality of modules, a first group within the plurality of groups corresponds to the first module, a first group input data within the input data is assigned to the first group, part of the first group input data is processed by the first module, and part of the first group input data is processed by the second module.
5. The acceleration system of claim 4 , wherein a processing speed of the first module is faster than a processing speed of the second module.
6. The acceleration system of claim 1 , wherein the acceleration system is used to accelerate processing of an input data, an input of the recurrent structural causal model is related to the input data at a first time point, the recurrent structural causal model outputs information corresponding to a second time point, the information comprises the numbers, hierarchies, or types of the plurality of modules, and the acceleration system is dynamically configured according to the information.
7. The acceleration system of claim 6 , wherein the input of the recurrent structural causal model is a time series data, the time series data comprises the input data at the first time point and a third input data at a third time point, and the third time point is earlier than the first time point.
8. The acceleration system of claim 7 , wherein the second time point is later than the first time point.
9. The acceleration system of claim 1 , wherein a causal graph is generated and verified with a causal discovery algorithm, and the recurrent structural causal model is trained using the verified causal graph.
10. A dynamic configuration method for an acceleration system, comprising:
providing a first information from a recurrent structural causal model based on a first series data, wherein the first series data comprises a first input data at a first time point; and
configuring numbers, hierarchies, or types of the plurality of modules of the acceleration system according to the first information to accelerate processing of a second input data at a second time point.
11. The dynamic configuration method according to claim 10 , wherein
each of a plurality of modules of the acceleration system comprises at least one central processing unit, at least one graphics processing unit, at least one field programmable gate array, or at least one application specific integrated circuit,
wherein at least one of the plurality of modules comprises at least another of the plurality of modules, such that the acceleration system is structured and nested.
12. The dynamic configuration method of claim 10 , wherein the second input data is divided into a plurality of groups, the plurality of groups correspond to the plurality of modules of the acceleration system respectively, and the plurality of modules of the acceleration system process the plurality of groups corresponding to the plurality of modules respectively.
13. The dynamic configuration method of claim 12 , the second input data is divided into the plurality of groups according to a clustering algorithm.
14. The dynamic configuration method of claim 12 , wherein a first module within the plurality of modules comprises a second module within the plurality of modules, a first group within the plurality of groups corresponds to the first module, a first group input data within the second input data is assigned to the first group, part of the first group input data is processed by the first module, and part of the first group input data is processed by the second module.
15. The dynamic configuration method of claim 14 , wherein a processing speed of the first module is faster than a processing speed of the second module.
16. The dynamic configuration method as described in claim 10 , further comprising:
providing a second information from the recurrent structural causal model based on a second series data, wherein the second series data comprises the second input data at the second time point; and
configuring the numbers, hierarchies, or types of the plurality of modules of the acceleration system according to the second information to accelerate processing of a fourth input data at a fourth time point.
17. The dynamic configuration method of claim 16 , wherein the first series data comprises the first input data at the first time point and a third input data at a third time point, the second series data comprises the first input data at the first time point and the second input data at the second time point, and the third time point is earlier than the first time point.
18. The dynamic configuration method of claim 16 , wherein the first information comprises the numbers, hierarchies, or types of the plurality of modules corresponding to the second time point, and the second information comprises the numbers, hierarchies, or types of the plurality of modules corresponding to the fourth time point.
19. The dynamic configuration method of claim 10 , wherein a causal graph is generated and verified with a causal discovery algorithm, and the recurrent structural causal model is trained using the verified causal graph.