Nengo's Hypotheses and Other Simulators' Frameworks

I am reading a lot about Nengo and it is clear that it is based on the Neural Engineering Framework (NEF) and also on the Semantic Pointer Architecture (SPA), both of which are, at this point, hypotheses.

What will happen in case some of the assumptions in NEF or SPA prove to be false? What will happen to Nengo as a tool and to NEF or SPA as frameworks? Can they be adapted radically in case their hypotheses are false? Also given that Nengo is used to build scientific models, which are usually published and new research is constantly based on top of the previous published work, what would a false assumption on the NEF/SPA mean for the validity of this body of published work?

One more question that naturally stems from the previous one has to do with the other available neural simulators (NEST, BRIAN, NEURON etc…). In most publications about Nengo it is mentioned that it is based on the NEF and SPA frameworks, but I have never managed to find any information regarding the frameworks the other simulators use to build their models. Where are those simulators based on?

PS: I am not sure if SPA is a framework that is used to build Nengo or if it is also built using the NEF .

Nengo is simply a tool for building and simulating neural networks. Building networks with the NEF is one of the things you can do with Nengo, or building SPA-style networks is another thing you can do with Nengo. Or you can build deep networks, or Hopfield networks, or do STDP experiments, or build a robotic arm controller, or pretty much anything that involves using neurons and connections between neurons to perform some function. Nengo is not tied to any of the assumptions of the applications that are built using that tool. In the same way, for example, that doing mathematical research with a hypothesis and finding that your ideas were wrong does not mean there is a problem with your calculator. Or if someone is doing research in deep learning and using TensorFlow, if their research hypothesis turns out to be false, that doesn’t really affect TensorFlow at all.

I don’t believe that NEST/BRIAN/NEURON have any specific theoretical framework associated with them (although I am not an expert in those tools). Like Nengo, they are simply tools for simulating neural networks (although different frameworks will make it easier or more difficult to build different styles of model).

SPA is a framework built using Nengo, rather than the other way around.

1 Like

Thank you for your specific answers.

Coming from the perspective of the NEF and the SPA as theoretical frameworks based on scientific hypotheses, several papers have been published recently that extend their foundations more broadly (*):

This means that if additional biological details turn out to have some functional significance — or relevance to improving computational trade-offs between precision, energy, and timing — then it is often possible to incorporate those details into existing models. NengoBio is an experimental research tool that works in this direction, based on the last paper from the list above.

(*) Non-exhaustive list, in no particular order.