Extended Data Fig. 6: Computation flow diagram of a Network of ENUs.
From: Network of evolvable neural units can learn synaptic learning rules and spiking dynamics
Shows a computation example with 4 ENU synapses and 2 ENU neurons, each having 3 channels. The sensory input neurons X are concatenated with all the ENU neurons H to get our input batch. A connection matrix is then applied that broadcasts (copies) the neurons’ output to each connected synapse ENU (1). On this resulting matrix we can then apply standard matrix multiplication and compute our synapse ENUs output in parallel (2). We can reshape this and sum along the first axis, as we have the same number of synapses for each neuron (3). This gives us the integrated synaptic input to each neuron ENU (4). Finally, we apply the neuron ENUs on this summated batch and obtain the output for each neuron in the ENU network (5). Each ENU has multiple outputs, so we have multiple channels that are processed by the ENU (the columns of each matrix), and we also have multiple neuron and synapse ENUs computed in parallel (the rows of each matrix).
