Really cool, and although I am not quite up to date on biology research (relevant here), I believe there were exciting results showing mechanisms like electric fields acting as signals for cells to differentiate in some species, i.e. "telling a cell where it is", and supposedly guiding "what it should do/become", etc.. I believe this was a result from studying Axolotls, the amazing self-regenerating (and severely threatened of extinction in nature) salamanders[1].
Side note: if we needed more reasons to conserve the amazing and enormous spectrum of life, one more reason is this kind of discovery that might enable better understanding (and maybe enhancement one day) of cell growth and regeneration in humans. Also showing that biology in many ways is extremely far ahead of what humans can achieve with current technology or will for the foreseeable future (as much as the automata example is very neat, it's nowhere near self-assembling full working and self-reproducing creatures from a compact genetic code!).
It seems you can donate directly to help Axolotl conservation (which again is critically endangered), seems really important if you can help! [2] (although there are of course many other means to help if you're interested in conservation in general!)
For a split second there I believed there is a new Distill publication! Their articles were the most inspirational and eye-opening resource on my beginnings of ML journey, the quality of visualizations definitely made lasting impact on my mental models of _what is going on_.
My favourite goes definitely to The Building Blocks of Interpretability (https://distill.pub/2018/building-blocks/), those images landed in a lot of my university presentations and the dog made everyone immediately interested ;-)
There is ongoing research on neural cellular automata, as they seem to be a very efficient way to generate pretraining tokens: https://arxiv.org/html/2603.10055v1
This has been shared a few times here. It looks cool but I don't understand the usefulness of this tech. This is likely a demo to show some capability, what that is though I don't understand.
I think there's utility in structures/patterns that can self-heal when damaged, but my imagination initially goes to usage in games and unusual UI.
I've played with reaction-diffusion quite a bit, which is an adjacent idea, and there can be some fun in trying to delete all the solitons - which can be an interesting challenge when they behave like like rapidly-replicating gliders!: https://www.mrob.com/pub/comp/xmorphia/F180/F180-k530.html.
My kids enjoyed that as a semi-game like challenge in an RD I made in Android app 'Shader-Editor' on phone, and I reckon something similar but using Neural-CA could be really fun!
Particularly if animated/dynamic (moving) patterns could be developed - like 'The Powder Toy' (et al) but less discrete and more continuous - and far less aliasing!
How about a Terraria-like crafting game, crossed with Powder Toy and using NeuralCA, Reaction-Diffusion and other continuous automata like fluid Sims too! Sounds like it could be fun!?
well it attempts to explain how individual cells can combine to create complex creatures (like a lizard) AND how the creature can have features like healing, regeneration, etc.
- is that a blog that you are familiar with that teaches you neural networks in the most animated visual manner that absolutely blows your mind off?
- because it finally clicks and you realize how it works on the inside?
- https://visualrambling.space/neural-network/
- i found one but it doesn't have text content explaining the math, the reason behind gradient descent, the need for a partial derivative etc.
- I would super appreciate if someone has a resource for cnn, rnn, and other types of neural networks like this
Really cool, and although I am not quite up to date on biology research (relevant here), I believe there were exciting results showing mechanisms like electric fields acting as signals for cells to differentiate in some species, i.e. "telling a cell where it is", and supposedly guiding "what it should do/become", etc.. I believe this was a result from studying Axolotls, the amazing self-regenerating (and severely threatened of extinction in nature) salamanders[1].
Side note: if we needed more reasons to conserve the amazing and enormous spectrum of life, one more reason is this kind of discovery that might enable better understanding (and maybe enhancement one day) of cell growth and regeneration in humans. Also showing that biology in many ways is extremely far ahead of what humans can achieve with current technology or will for the foreseeable future (as much as the automata example is very neat, it's nowhere near self-assembling full working and self-reproducing creatures from a compact genetic code!).
It seems you can donate directly to help Axolotl conservation (which again is critically endangered), seems really important if you can help! [2] (although there are of course many other means to help if you're interested in conservation in general!)
[1] https://youtu.be/7cLaU_agj6k?&t=86
[2] https://www.moja.ong/programs/axolotl-habitat-conservation/ https://www.moja.ong/donar/
For a split second there I believed there is a new Distill publication! Their articles were the most inspirational and eye-opening resource on my beginnings of ML journey, the quality of visualizations definitely made lasting impact on my mental models of _what is going on_.
My favourite goes definitely to The Building Blocks of Interpretability (https://distill.pub/2018/building-blocks/), those images landed in a lot of my university presentations and the dog made everyone immediately interested ;-)
There is ongoing research on neural cellular automata, as they seem to be a very efficient way to generate pretraining tokens: https://arxiv.org/html/2603.10055v1
Morphogenesis data compression is pretty impressive, the human genome is only ~700Mb of data.
Ahh but it requires the laws of physics as a runtime.
This has been shared a few times here. It looks cool but I don't understand the usefulness of this tech. This is likely a demo to show some capability, what that is though I don't understand.
I think there's utility in structures/patterns that can self-heal when damaged, but my imagination initially goes to usage in games and unusual UI.
I've played with reaction-diffusion quite a bit, which is an adjacent idea, and there can be some fun in trying to delete all the solitons - which can be an interesting challenge when they behave like like rapidly-replicating gliders!: https://www.mrob.com/pub/comp/xmorphia/F180/F180-k530.html.
My kids enjoyed that as a semi-game like challenge in an RD I made in Android app 'Shader-Editor' on phone, and I reckon something similar but using Neural-CA could be really fun!
Particularly if animated/dynamic (moving) patterns could be developed - like 'The Powder Toy' (et al) but less discrete and more continuous - and far less aliasing!
How about a Terraria-like crafting game, crossed with Powder Toy and using NeuralCA, Reaction-Diffusion and other continuous automata like fluid Sims too! Sounds like it could be fun!?
well it attempts to explain how individual cells can combine to create complex creatures (like a lizard) AND how the creature can have features like healing, regeneration, etc.