Fast learning without synaptic plasticity in spiking neural networks

This system involves having a spike network with neuroplasticity that is trained on adjusting the underlying network to succeed at exams. They specifically do not give the working neurons plasticity parameters as the goal is to determine if the overseer network alone is enough to enable rapid adjustment to problems.

The overseer network essentially learns the goals of ex. moving an armature with different mass and length while the control network is randomly initialized. The overseer network is able to correct the control networks to account for dynamic changes in weight of an armature within five time steps.

Quinn note: you basically need a network that knows what the intended outcome is, and it modifies the control system based on this. In this case the desired results were fixed by the lab. For a self-directing system we have to ask if its enough to simply predict the future and use the misprediction as a signal that the control system is out of alignment?

Biological data suggest that a synergy of synaptic plasticity on a slow time scale with network dynamics on a faster time scale is responsible for fast learning capabilities of the brain.

We show here that a suitable orchestration of this synergy between synaptic plasticity and network dynamics does in fact reproduce fast learning capabilities of generic recurrent networks of spiking neurons.

However recently it has been shown that a substantial fraction of the functional capability of networks of LSTM units can be reproduced by networks of spiking neurons, provided that they also contain neurons with spike frequency adaptation (SFA)10,11.

SFA means that a neuron increases its firing threshold after firing.

Experimental data from the Allen Institute19 show that a substantial fraction of excitatory neurons of the neocortex, ranging from in mouse visual cortex to in the human frontal lobe, exhibit SFA. Therefore, we endow only a subset of spiking neurons with SFA in line with this data (Also see Fig. 8 in Salaj et al.11 for a plot of the distribution).