Comparison of Neural Network Simulators (survey)

From emergent
Jump to: navigation, search

Back to Comparison of Neural Network Simulators.


Survey

We are working on a survey to send out to the authors of every simulator in the comparison. This provides the authors an opportunity to provide clarification about the intended role of the simulator, and it allows users to compare the responses side by side. All of these questions have been user contributed, feel free to add yours!

Instrumentation

Having implemented a network, how easy is it (or what facilities exist to) after the fact to place measurement 'probes' on anything that might conceivably be measured (from single values to a whole vector of such values across units) and have their histories captured in a format that is conducive to further analysis?

Experimentation

Having implemented a network, how easy is it (or what facilities exist) to program an experiment to put the network through it's paces (e.g. pick the next stimulus from a file; present the stimulus; hold the stimulus until some measurement stabilizes; iterate over stimuli until weights stabilize; iterate over probe stimuli)?

Size Constraints

How many neurons and synapses will can be simulated on a standard mid-range computer? How many neurons and synapses can be simulated in real time on a standard mid-range computer? How accurate, or what is the value of a single "time step", in the simulations? (The last two questions may not apply depending on the model used).