Thanks everyone for watching! Let me know what you think or what you'd like to see in the future, in the comments below! Subscribe to support the channel :)
I think part of the problem is that Game of Life Rules are chaotic, in the sense that a small change to the rules will completely change how the final thing behaves. One thing you could change (which would drastically alter the nature of the project but which I find interesting) would be to use a probabilistic cellular automaton. Something like “if I am alive and have 4 neighbors, I have a 30% chance of staying alive and a 70% chance of dying.” To keep stuff from growing out of nowhere, you would say that anything with 0 neighbors always dies/stays dead, but then you have 16 parameters to tune for 1-8 neighbors for both alive and dead
9:24. OMG, this pattern must follow the golden ratio (which allows the most chaotic structures in nature). It looks like seeds in a sunflower so much. For more info check the video by Numberphile called "The Golden Ratio (why it is so irrational) - Numberphile"
It'd be interesting to see this with a hexagonal grid Weirdly would have less neighbors for each cell unless you included the "diagonals" of the hexagon Same with a triangle grid, seeing how all of those compare
Cool idea! Found a paper doing something similar: link.springer.com/referenceworkentry/10.1007/978-1-4939-8700-9_58 -- definitely something worth trying
This is incredibly well made! One thing I think you could work on is the audio because you can hear the cuts in some places, but other than that very interesting concept and very interesting application!
@@maximejkb If he meant how AI can be applied in data compression, one way that I know of is video compression. Right now RUclips is adopting AV1 video codec, but on my research on it I found it frustrating that they didn't include AI based upscaling technology. For example you can take a source 1080p video, using a video codec compress it and downscale to 720p (roughly 50% of total pixel count, and weirdly enough even lossy video compression scales close to linear) and put an upscaler layer on top of the video decoder (FSR or DLSS equivalent). I think as of right now the usage of AI for video and data compression is limited, for video maybe macroblock placement uses something simillar to AI. There are however lossless data compression algorithm called CMIX which might use something more related.
@@montec6113 Woah super cool to know -- I'm also super surprised to hear there are people on RUclips interested in learning about compression! My first ever paper was on AI-based image compression: arxiv.org/abs/2202.00198
@@maximejkb Oh wow, now I feel a bit silly being an amateur trying to explain these things to an expert lol Skimming your paper and it seems that you already explored these ideas of improving PSNR scores with AI upscaling? (I need a bit more time to fully figure out the methodology that goes in there). Also as a side note, I hate SSIM with passion, from my extensive encoder benchmarking it would not provide a reliable objective score on how a video looks and PSNR in this regard was a lot more sensitive and useful, however it has issues with noise in the videos/images. As you probably already know encoders try to reduce entropy and denoise the video, this dramatically affects PSNR scores, but in real subjective viewing the quality is higher than we might assume judging from the score. Right now the state of the art measuring tool for video quality is called VMAF, it's been developed by Netflix. In my experience score of 97 is more or less visually lossless (it's not, but our eyes aren't good enough to see the difference) and anything above 95 is deemed to be of acceptable quality. I recommend giving it a shot, because VMAF is I think a neural net based avaluation tool which was trained to score videos as close to human vision as possible.
@@maximejkb Nah, I mean that if you haven't tried your hand of data compression yet, you'll very likely enjoy that as the next project. It has some similarities and some differences to the cellular automata area!
If you are interested, there is a program called DarwinBots 2 I used to play around with it and 3DVCE It is very interesting to see how "life" emerges and finds solutions Great video!
That was from an early version of this project -- basically where I was letting each cell see a set of concentric circles around itself. It's one of the most interesting patterns I've seen so far, so I want to try to recreate it in a more principle way!
Entropy in computer science means some thing slightly different from in physics / chemistry - basically you measure it as the number of 0/1 bits it would take to represent the distribution of values. ruclips.net/video/v68zYyaEmEA/видео.html has a really good explanation
Thanks everyone for watching! Let me know what you think or what you'd like to see in the future, in the comments below! Subscribe to support the channel :)
This was really cool, would love to see more of it!
Thanks for watching! Will be working on this in the background!
This is really interesting and fun to watch! Hope you become more well-known.
Appreciate that! Thanks for watching
I think part of the problem is that Game of Life Rules are chaotic, in the sense that a small change to the rules will completely change how the final thing behaves. One thing you could change (which would drastically alter the nature of the project but which I find interesting) would be to use a probabilistic cellular automaton. Something like “if I am alive and have 4 neighbors, I have a 30% chance of staying alive and a 70% chance of dying.”
To keep stuff from growing out of nowhere, you would say that anything with 0 neighbors always dies/stays dead, but then you have 16 parameters to tune for 1-8 neighbors for both alive and dead
Oh I like that -- that's super interesting. I'll play with that and see if it yields anything cool.
9:24. OMG, this pattern must follow the golden ratio (which allows the most chaotic structures in nature). It looks like seeds in a sunflower so much. For more info check the video by Numberphile called "The Golden Ratio (why it is so irrational) - Numberphile"
It'd be interesting to see this with a hexagonal grid
Weirdly would have less neighbors for each cell unless you included the "diagonals" of the hexagon
Same with a triangle grid, seeing how all of those compare
Cool idea! Found a paper doing something similar: link.springer.com/referenceworkentry/10.1007/978-1-4939-8700-9_58 -- definitely something worth trying
I'm here for the maths, this is good stuff! Beyondway!
Appreciate that! Thanks for watching!
This is incredibly well made! One thing I think you could work on is the audio because you can hear the cuts in some places, but other than that very interesting concept and very interesting application!
Appreciate that feedback! You mean cuts between dialogue?
I remember having tons of fun with these!
Next up: Data compression?
They're super cool! And you mean you'd be interested in a video on compression -- in any specific area?
@@maximejkb
If he meant how AI can be applied in data compression, one way that I know of is video compression. Right now RUclips is adopting AV1 video codec, but on my research on it I found it frustrating that they didn't include AI based upscaling technology. For example you can take a source 1080p video, using a video codec compress it and downscale to 720p (roughly 50% of total pixel count, and weirdly enough even lossy video compression scales close to linear) and put an upscaler layer on top of the video decoder (FSR or DLSS equivalent). I think as of right now the usage of AI for video and data compression is limited, for video maybe macroblock placement uses something simillar to AI. There are however lossless data compression algorithm called CMIX which might use something more related.
@@montec6113 Woah super cool to know -- I'm also super surprised to hear there are people on RUclips interested in learning about compression! My first ever paper was on AI-based image compression: arxiv.org/abs/2202.00198
@@maximejkb Oh wow, now I feel a bit silly being an amateur trying to explain these things to an expert lol
Skimming your paper and it seems that you already explored these ideas of improving PSNR scores with AI upscaling? (I need a bit more time to fully figure out the methodology that goes in there). Also as a side note, I hate SSIM with passion, from my extensive encoder benchmarking it would not provide a reliable objective score on how a video looks and PSNR in this regard was a lot more sensitive and useful, however it has issues with noise in the videos/images. As you probably already know encoders try to reduce entropy and denoise the video, this dramatically affects PSNR scores, but in real subjective viewing the quality is higher than we might assume judging from the score. Right now the state of the art measuring tool for video quality is called VMAF, it's been developed by Netflix. In my experience score of 97 is more or less visually lossless (it's not, but our eyes aren't good enough to see the difference) and anything above 95 is deemed to be of acceptable quality. I recommend giving it a shot, because VMAF is I think a neural net based avaluation tool which was trained to score videos as close to human vision as possible.
@@maximejkb Nah, I mean that if you haven't tried your hand of data compression yet, you'll very likely enjoy that as the next project. It has some similarities and some differences to the cellular automata area!
If you are interested, there is a program called DarwinBots 2
I used to play around with it and 3DVCE
It is very interesting to see how "life" emerges and finds solutions
Great video!
Oh no way these are so cool! Ok I definitely want to play around with them, thanks for the idea!
Very well edited. Hope this gets picked up by the algorithm! Remember me when you get that diamond play button 😁
Haha appreciate that!
what was the pattern for this one at 9:25? it looks really interesting im curious
That was from an early version of this project -- basically where I was letting each cell see a set of concentric circles around itself. It's one of the most interesting patterns I've seen so far, so I want to try to recreate it in a more principle way!
Could someone enlighten me: how do you measure "randomness" and "entropy" from 0s and 1s in a grid? Thanks
Entropy in computer science means some thing slightly different from in physics / chemistry - basically you measure it as the number of 0/1 bits it would take to represent the distribution of values. ruclips.net/video/v68zYyaEmEA/видео.html has a really good explanation
so do you code in vim now?
HAHA lowkey only for the aesthetic... or when coding over SSH on a robot