Algorithm vs Execution Time discussion small
Today for small discussion of algorithm vs Execution time.
And there are small discussion for some types of executions.
Battery SOC calculation algorithm
it is derived as following paramters:
For above calculas, the integral method is in essence using the multiply accumulation methodology:
from t-->t+deltat:
i*t-->(i+deltai)*(t+deltat)
And then add up, so it is the multiply and addition combination.
(1) multiply
(i+deltai)*(t+deltat)
(2) add up
...+(i+deltai)*(t+deltat)+(i+2*deltai)*(t+2*deltat)+...
Calculation time is O(nlogn), similar to following:
void eta1=const
void eta2=const
void C=const
void accumulate_soc
double t0=0.00
double i0=0.00
While (t0<=t<=t1)
{
def i=i0
t=t0+0.0;1i_t=fun(i0,t)
delta_soc_delta_t=(eta1)*(i+deltai)*(t+deltat)/((eta2)*C)
accumulate_soc=zigma_t0-->t_(delta_soc_delta_t)
}
so the average time add up against n variables putting add up, the time is nlogn
2.Covolutional Neural network
it is the matrix to be calculated:
X*theta=Y
X is m*n matrix, theta is n*n covolutional quotient, then Y is the m*n matrix.
And then the Covolutional again and again in the photo, for the process:
n times of the addition in X
m*n times of the multiply in theta and X
then it is mainly equal to n^2 in the same level
so the average time add up against n variables putting add up, the time is O(n^2)
3.Relationship between algorithm and time cost
Source from reference.
4. CPU speed, and execution speed regarding different algorithm and time cost
Three points will impact huge to AI executing speed:
Algorithm choice
Coding language choice
Hardware chip choice
Main problem in the Algorithm is to slashing down the cost and refer to the themes we had discussed before.
Thanks for your time!
[Reference]
https://m.toutiaocdn.com/item/6814334289279189511/?app=news_article×tamp=1588204983&req_id=20200430080303010014051081071951C1&group_id=6814334289279189511
- 用户评论