Reading the launch
PC-ALM is basically asking a simple question: why do neural networks need backprop if the brain never had it? it trains deep networks through local dynamics instead of global gradients, and somehow scales to 1000 layers. local learning, no backprop. maybe
Reading trades.