This doesn't look like a neural network that's solving any actual problems beyond visualizing firing patterns. Most ANNs don't have any concept of time built into the model. When you're in feed-forward mode, the underlying computation is simply a bunch of dot products.
However, you might be able to make some cool visualizations if your ANN was a Spiking Neural Network.
Well it's thresholded so you could visualize the "activated" neurons in a manner such as this.
Would it be useful though? Maybe. Probably not as the fact that a neuron fires isn't nearly as useful as why it fired (what it represents).
But I like that you mention time. It always seemed odd that NN's today completely ignore that aspect. Sure whether a neuron fires is binary. But the accumulation of spikes is not binary and highly temporal. yet we completely ignore this aspect.
I'm confused by you guys saying NN's today completely ignore time? If I have a network of Hodgkin-Huxley neurons and I integrate the equations, I'm integrating over time. I'm not sure how that is ignoring it? Disclaimer: it has been about 4 years since I've written any code to integrate H-H neurons, so I might be forgetting something really obvious. :) But I can't imagine EVERYONE is ignoring time when they simulate networks of neurons.
Is it based on an actual model of neural activity (Izhikevich, H-H, etc.) or is it an abstract visualisation in the sense that it just looks like neurons firing?
There is a serious amount of heavy lifting in the visualization department done here for us. I imagine forking this repo and adding a simple model to govern the neuron and axon behavior would be doable. Writing this off as "pretty pictures" is a little over-the-top. Seems like a significant amount of work went into producing these real-time visualizations.
Shameless plug for a little visualization I made some time ago of a neural network in a developing fruit fly, which performs an interesting probabilistic algorithm to elect a maximal independent set http://jberryman.github.io/fly-mis/
It somewhat depends on what you think of as a neural network. For simple models no, state is vary important, as is signaling delays, hormones, and blood flow etc. Which means each neuron would individual be far more complex than what is generally used in computer science. Further brains are far from random so you would need to either copy someone or simulate the various developmental stages a person goes though. Finally, a neural network is pointless without some output so for anything beyond pretty pictures you need some sensory input's and some way to handle outputs.
However, I have seen estimates that somewhere around a yottahert class computer working for a few years might be capable of simulating a brain sufficiently complex to hold an arbitrary conversation. Building the software capable of running such a simulation is another story.
This looks absolutely stunning. I wish I could train the brain by myself and see how it operates in as an artificial neural network. Anyway, great job!
Everyone complaining or fussing about how primitive this model is on a browser, ya'll need to chill out and stop taking beautiful work like this for granted. Some of you may say this is easy to do or why didn't he also "do this" or "do that" - how about you help out and make it better ;)
I think the complaints are fair based on the title of the HN post which indicates that this is a visualization of a biological neural network, which it empathetically is not (it's not a true representation of any kind of neural network, at least not that I know of, but a visualization of graph traversal where the graph happens to look like a brain).
So, it is cool -- extremely pretty -- but fairly misleadingly named in my opinion. It could be more accurately titled, "An artist's rendition of a brain".
I'm not complaining about how primitive it is. I'm complaining about using buzzword when at the end it is just a random network disguised in a brain structure (and I'm just talking about geometry).
I agree, a really cool project which I would imagine could be extended to a "real" model without a ton of work. I imagine the amount of work to fork a javascript library which integrated a network of H-H model neurons to make it output this type of visualization would be far more work than adding the H-H piece to this code. I could be wrong but at first blush, that's how I see it.
I think it would be quite a bit of work to integrate the H-H model or any model of a biological brain with this visualization.
This is because the visualization is basically just a graph, with axons as edges and neurons as nodes, and then signals are propagated with extremely simple graph-traversal style logic. On the other hand, the H-H model tries to deal with the physical reality of the human brain so it is much more complicated and relies on some data (like axon width) that simply doesn't exist here.
It would be possible and potentially interesting, for sure, but probably more work than you might think.
