Ok but in order to understand how to do modification requires understanding the underlying theory. Let's say eg you want to change the learning rate α > Changing it can really improve your algorithm for your specific case? > Does your algorithm still converges to the real parameters? > The final estimator is still consistent? > What's the relationship between the learning rate and the speed of convergence?
Physics? No. You should understand vector calculus and linear algebra. Sadly, that filters out many of the CS tards, so if that includes you, then you need to go back to webdev and join pajeet.
I would focus more on multivariate statistics. Linear algebra per se is just a basis as would be calculus. There's no need to focus on linear algebra unless you are doing research
Don't be retarded. It does use basic calculus concepts (mse, derivative chain rule, gradient computation) but the concept of adaptive algorithm and its analisys (mixing time, convergence, stabilizing solutions, partial consistency, etc) are not undergrad concept
imagine using math while coding.
yes
yes or you cant have your shooting game with shit jump physics
>pubg
Yes and no. It's not like undergrad math class where you're grinding problem sets and crunching numbers, but you need to understand the math.
>you need to understand the math
just read and understand existing patterns/solutions only, or write and change too ?
That's great until you have to debug the autograd weirdly collpasing all your weights to zero.
Ok but in order to understand how to do modification requires understanding the underlying theory. Let's say eg you want to change the learning rate α
> Changing it can really improve your algorithm for your specific case?
> Does your algorithm still converges to the real parameters?
> The final estimator is still consistent?
> What's the relationship between the learning rate and the speed of convergence?
physics and neural nets (and video encoding) are all just signal processing. get comfy.
i need some fat fucking cocks in my butt lord im such a gay for bbc
this is what i pretend i'm doing when im merging models, training loras and hypernetworks and crafting prompts to get anime sluts
Physics? No. You should understand vector calculus and linear algebra. Sadly, that filters out many of the CS tards, so if that includes you, then you need to go back to webdev and join pajeet.
I would focus more on multivariate statistics. Linear algebra per se is just a basis as would be calculus. There's no need to focus on linear algebra unless you are doing research
i don't see any physics in that picture
>he can't mathematically model things and wants to specialize in mathematical models
NGMI
Physics? No
> Pic
Pic is just gradient computation and a stochastic gradient descent ( Hessian(loss function) approximated as α/2)
That shit is basic calculus and a week on the specific topic
this is what you show people who never got past hs math so that they think youre a god-tier genius
I'm an Industrial Engineer. That's some undergrad shit.
Don't be retarded. It does use basic calculus concepts (mse, derivative chain rule, gradient computation) but the concept of adaptive algorithm and its analisys (mixing time, convergence, stabilizing solutions, partial consistency, etc) are not undergrad concept
don't remember doing differentiation or series to this degree
no but you have to market yourself like you did