Machine Learning library comparisons for testing brain architectures

There are already nice ML comparisons out there:

Here, I compare ML libraries for suitability of use in archibrain.
Particularly, we want

1. arbitrary even spiking neurons
2. auto-differentiation of arbitrary functions
3. ease of implementing different learning algos (not just backprop) - almost Brian style.
4. can use GPUs and CPUs across multiple nodes.
5. ease of implementing both RL and supervised learning tasks.

library	arb nrns?	auto-diff?	arb learn algo?	multi-node GPU/CPU	tasks