On the other hand, neuromorphic architectures don’t have the same constraints. So why can’t we stick a capability like sampling of weights into there as well? In particular, in our COINFLIPS project we’re examining devices like magnetic tunnel junctions that can be a weighted coin. 3/
At first blush, this is a terrible idea. Try to Monte Carlo sample weights of any decent sized neural network on a GPU. Get back to me when you’re done running that… this is far from efficient on today’s systems. There is a reason most sampling on things like VAEs and dropout is on neurons. 2/
I’m going to kick off my Bluesky experience with a summary of a paper I’m presenting at the IEEE Rebooting Computing meeting this week. The premise is that the noise in our brain is in its synapses, so shouldn’t it make sense to sample neural networks the same way? arxiv.org/abs/2311.130...
Probabilistic artificial neural networks offer intriguing prospects for enabling the uncertainty of artificial intelligence methods to be described explicitly in their function; however, the...
Please add me! I'm a neuroscientist in exile doing brain-inspired computing research scholar.google.com/citations?us...