Improvement idea for Neuron group synchronization concerning "delay" & "output-recovery-time" signal firing performance

[learner411]

Hello. I love using Membrane as it is the most intuitive yet flexible and broadly capable neural network simulation, training and editor tool i've seen out there. I am so close to hitting a major milestone in a project I have been working on involving the use of RNN capabilities. as lovely as the feature "link" from the "edit link property" interface it. I notice although is a success in its use for other design/forms of time-variant neural networks. its behavioral properties. It only allow for individual links to be postponed (for arbitrary time, according to "length" parameter setting) from an update during pattern run-throughs while undergoing each single teach-step(training step). I doesn't however have the ability to temporally prevent/freeze any upstream connections (meaning: other pre-occurring neurons and their links, closer to the input, that are connected) from being influenced by updates during teach-step (run-through of the patterns though a lesson course). Rather it can only prevent/freeze only its own update. I propose and insist that it a good idea to extend the capability of delaying firing of signals in a manner that allows a selected group (so that means both all the upstream and their downstream neurons all alike) of neurons entirely to all undergo having the ability to temporarily freeze during update by teacher, according to the timing setting of their delay link "length" and "output-recovery-time". so that they all fire in harmony as though they are one entity/unit.

I imaging similar to how convolution of links values works by in a sense merging links together. likewise there could a bottom interface to merge the fire in unison of a selected and group of neuron who are all assigned the all share that affiliation with each other.

I hope you can understand how this will be different from how the regular link "length" and "output-recovery-time" mange to work in principle. try as one might this cannot be used make a group of neurons that all have their delay length time synchronized and their signal firing in harmony (harmony is almost as to say: instead of firing sequentially the fire in parallel, as if all neuron behave like one single neuron where delay firing is concerned).

Also if possible can the activation-threshold also be made to obey the freeze from update rule too not just the links of the group of neurons. this is kind of a must do mandate if allowed or possible.

I've seen the magic you've accomplished and achieved by making the smart convolution of links in the near past already. it was fantastic I must say. I feel a great chance is there for you to pull this next one off, whomever this post may concern, I which case I would be speaking to "Thomas Jetter". There is not other like MemBrain out there as far as I can see

Thanks for hearing me out if so Thomas Jetter.