I'd like to see manufacturers develop a fiber optic based neural microchip. I've seen fiber optic chips and neural chips.. but why has nobody thought to combine them? Simulators such as this are all fine and good, but until somebody actually develops a truly sophisticated parallel processing neural network, AI really cant get very far.
How complex would an artificial neural network have to be to play a highly complex game with human players like Halo 3 or Grand theft auto 4? There are AI elements built into these games to play against humans, however, I'd like to see something akin to chess match between Deep Blue and Kasparov but for the Iron Man of Gaming instead of chess. There would be a huge market for highly challenging video game AIs.
Nice video, if you don't mind a couple of questions. What neural network solution did you use, how dense was the network, what were your inputs and outputs and how much better did it get with further generations?
I wrote a custom solution for this problem; I believe the agents in the video had 160 inputs, representing different subsections of the playing field, and one representing the number of junk puyos about to fall. The output is the perceived value of the field. You can see it simulating the effect of placing its next piece; it selects the move with the highest value. The project was never completed, so there are a number of things that keep the agents from improving much beyond what you see here.
How many generations have you seen it go to, and was there notable improvement?
mutehero7 10 months ago
any chance of you releasing the source code to the public? i'd love to learn how it works!
brawllegend 1 year ago
how do they become better at it? are u able to let them actually self improve? or do you need to set the paramiters/algorythms/wtv?
monoham1 2 years ago
I'd like to see manufacturers develop a fiber optic based neural microchip. I've seen fiber optic chips and neural chips.. but why has nobody thought to combine them? Simulators such as this are all fine and good, but until somebody actually develops a truly sophisticated parallel processing neural network, AI really cant get very far.
LordNexii 3 years ago
How complex would an artificial neural network have to be to play a highly complex game with human players like Halo 3 or Grand theft auto 4? There are AI elements built into these games to play against humans, however, I'd like to see something akin to chess match between Deep Blue and Kasparov but for the Iron Man of Gaming instead of chess. There would be a huge market for highly challenging video game AIs.
DeimosSaturn 3 years ago
The "AI" is taylored for the specific game. They don't "learn" as the AI here does.
SlykeThePhoxenix 2 years ago
Nice video, if you don't mind a couple of questions. What neural network solution did you use, how dense was the network, what were your inputs and outputs and how much better did it get with further generations?
FrankDodd 3 years ago
I wrote a custom solution for this problem; I believe the agents in the video had 160 inputs, representing different subsections of the playing field, and one representing the number of junk puyos about to fall. The output is the perceived value of the field. You can see it simulating the effect of placing its next piece; it selects the move with the highest value. The project was never completed, so there are a number of things that keep the agents from improving much beyond what you see here.
BugCatcher255 3 years ago