Machine Learning Control: Genetic Algorithms
HTML-код
- Опубликовано: 7 июл 2024
- This lecture provides an overview of genetic algorithms, which can be used to tune the parameters of a control law.
Machine Learning Control
T. Duriez, S. L. Brunton, and B. R. Noack
www.springer.com/us/book/9783...
Closed-Loop Turbulence Control: Progress and Challenges
S. L. Brunton and B. R. Noack
appliedmechanicsreviews.asmedi...
www.eigensteve.com/
This video was produced at the University of Washington 
Наука
I subscribed after watching about 12 seconds of your explanation! You are doing a very good job, and I don't know why is such a good content is so underrated! Please keep up the good work, you are making tomorrow's world a better place! Thank you
never tried understanding the concept of GA because it seemed complicated. but after watching this, I at least have the feel for it. great lecture
It is the perfect video I have ever watched on RUclips! It is a clear illustration. I admired the point you perfectly make a connection between tuning a PID controller and GA. Thanks a lot
Imagine being within the elite category of elite genetics and have your spouse delay procreation, would it be logical to marry such an individual, would such an replication be sustainable? Please advise.. I am the volition that an crossover would supersede an replication.
Thanks Dr. Brunton, it was a great illustration!
from this i understood mutation and crossover( Explore and exploit). thanks for the great objective lecture
incredible explanation!
This is *really* well explained. Thank you!
You're very welcome!
@Byron Yahir Why are you gae?
@Alonzo Cameron You are gae!
great intro, the sky is the limit!
Amazing!
One problem with tuning PID here is that it is difficult to evaluate the cost function automatically , the motor could be damaged or else if you just use random mutations on control laws. Also i think one of the problems with Genetic algorithms is that it is not really clear how the binary encoding of the parameter space works, I think these concepts easily port over to Evolutionary Strategies or ES by Schwefel and Rechenberg, where this is much more straight forward.
Based on the diagram shown at 11:00, I'm a little confused on how the "probability of selection" (or the second coin flip, as you say in the video) influences the GA. It seems like all of the individual parameter samples shown make it through the evolution from generation k to generation k+1. Does generation k actually have many more individuals than generation k+1? This may be covered in a later video...
Amazing
Hi, i love your work, what would you use to tune pid if not genetic algorithms for unknown system dynamics?
When crossing-over, how does the algorithm know which genes are the cause of a favorable trait? It seems exploitative relative to mutations but I don't understand the mechanism other than an average of a better pool.
Random scrolling lead me to learn something new. Thank you! I have two questions, would be great if I get answers.
• Can we say that Elitism and Replication are same, and can be duplicate in Generation(k+1)?
• Is there only two Generation(k) and Generation(k+1), or they can repeat until it find best parameters?
Once again thank you!
At 12:06 the exact moment you notice Cross Over is not when you are saying .. 'seen ( the eyes go on the left the voice slows down ) befooooore'
Since dimensiality reduction is supposed to be the main objective when you start, why would you worry about using an algo that performs well in High Dimensionality ? Thanks
Hi Steve, just wondering how can we encode continuous variables into gene sequences in GA? Discretization? Will this hurt the performance?
It is actually very easy I have already done that. Look at google as real value mutation and crossover
Good job. Well explained. Do u have a video of the next part ? (Matlab implementation)
Thanks! Yes, here are the next two videos: ruclips.net/video/S5C_z1nVaSg/видео.html and ruclips.net/video/Idl2wlnpDHU/видео.html
@@Eigensteve thank you so much Pr Steve Brunton
Thanks for the great video. However, I cant understand our Kp, Ki and Kd are functions of time or not. I mean control parameters change over time or we tune them and then they are constant. The other question about how we choose the cost function??
hey i'm 11
i think that every thing is possible nevertheless how old they are. age is just a number
when you lose it's not over , but if you quit it's over😁😉