Intel Scales Neuromorphic Research System to 100M Neurons

newsroom.intel.com

2 points·joubert··4 comments

Why does this bullshit keep being posted here? Zero reasons for this line of research to exist.

Why is this a waste of time? I know nothing about it but it seems interesting enough.

Because the only reason for why they built it is "let's build something that kinda looks like a brain!". Two problems with that though:

1. It's nothing like a real brain, because we don't really know how a brain works (if we knew, we would have simulated it in software long time ago). Human Brain Project has failed spectacularly with far more money and brain power pumped into it, but at least it advanced neuroscience in some ways (hopefully). This project advances nothing.

2. There are no useful applications these spiking networks implemented in hardware can do better than regular GPUs running regular deep learning models, and no reasons to believe this will change any time soon.

Someone at Intel is trying to make a career out of it, and a bunch of second grade researchers are trying to get grant money while the 'neuromorphic' buzzword is still hot.

After looking into it, it does seem that it's simply a hardware implementation of SNN's. And reading from the wikipedia, it does seem that SNN's are less hardware intensive anyway so I am now wondering why make hardware specifically for it if, as you said, GPU/CPUs can run it no problem. I'm also wondering why they would double down on hardware when from cursory research there isn't a holy grail of training like there is in ANN's.

Still, looking at SNN's was pretty interesting.

Reply on news.ycombinator.com