This is just the brute force solution. I'm pretty sure that's no more efficient than guessing all pixels at once and trying to check of that worked or not.
It's the trivial solution. It's significantly more efficient than using backprop. It's much simpler and leads to practically the same result.
This makes me wonder why you would use backprop to do this. It feels like a waste of time, and I'm not sure why this post generates so much attention on HN.
Also see https://nullprogram.com/blog/2020/11/24/
> I'm pretty sure that's no more efficient than guessing all pixels at once and trying to check of that worked or not.
It's iterative so you don't have to run for a very long time, and it will keep improving the loss function for longer.
It's true that a "good" decision now might turn out to be "bad" later, but whether it's effective in improving a solution depends on the fraction of times that happens, which is almost certainly not 50%. Hill-climbing methods like this are used everywhere in optimisation when you want a decent solution quickly and don't require optimality.
>somewhat analogous to predicting that a coin flip will land on 'heads' if it landed on heads at the last flip
I'm no statistician either, but this is not analogous at all.
I just mean with the known complexity and nature of Game of Life, trying to guess things based on an objective function, when going back in time, seems no more efficient than trying to guess coin flips, based on prior flips. There's probably some Logic Theorem (sorry I don't know it) that describes why objective functions cannot be used to help solve the backwards Game of Life problem.
It's a complexity problem like which butterfly "caused" a tornado, or which initial clumping of rocks "caused" a black hole, because there are infinite numbers of solutions. Like how many initial GoL boards are there that result in a particular final state? I say for any final state there will be infinite possible starting states that can lead up to it.