AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Achieve Near Linear Speed Up in Sparse Networks With Cerebrus
Cerebrus chip has 800 an 50 thousand cores and forty gigabites of a memory on board. Thee cerebrus cores never multiply by zero, so all the zeros are filtered out. This in turn provides a performance advantage by doing useful work during those cycles,. Not to mention the power and efficiency savings. But is there something fundamentally special about sparsity? Numenta certainly seem to think so. They anecdotally point to the brain as being sparse. But the brain is probably sparse for the same reason that i don't to get up every morning and travel to every city in the uk. Doesn't seem like a profound insight, honest.