/sci/ - Science & Math

Maths is a cumulative subject. I am picking it up again and falter (as I have before) around trigonometry (esp. trig. - God do I hate trig.), functions, and calculus. My route to learning these is usually scattered and inconsistent. Can anyone describe a linear learning route through these fundamentals to make the process any more straight forward? Seems like a mess of contingencies to me, but my IQ is probably very low.

>>11389019
a genetic algorithm would be better suited to designing analog circuits, since there is a high amount of intra-circuit-dependencies in the search space of circuits.
I think this is the property that RL has trouble dealing with because the output of an RL policy must be a sequence, where as genetic algorithms can use almost anything as a genome as long as you can mutate it and perform crossover. This allows them to represent complex structures like graphs; if you wanted to represent such a think with an RL policy i think you would have to require the RL policy spits out a lot of book keeping type stuff that may be harder to learn using gradient descent.

>>11387870
you are right, i think the search space in that paper for the RL policy is too small to really be comparable to the GA and too small to draw any conclusions.
Since you seem to be more knowledgeable with this stuff than I, I would be curious to know if you think RL can handle every problem that GAs can? Or are there search spaces where one might be better than the other?

>>11387929
Not related at all really. Genetic algorithms (GAs) are classical ML and simulate the process of natural selection, where as GANs are deep learning based (and pretty cool)