Gneural Network - Tasks: task #14197, Support Spiking Networks.
You are not allowed to post comments on this tracker with your current authentication level.
task #14197: Support Spiking Networks.
Submitter: | Ray Dillinger <rayd> | ||
Submitted: | Sun 30 Oct 2016 05:24:47 AM UTC | ||
Should Start On: | Sat 29 Oct 2016 07:00:00 AM UTC | Should be Finished on: | Sun 29 Oct 2017 07:00:00 AM UTC |
Category: | None | Priority: | 4 |
Status: | None | Privacy: | Public |
Assigned to: | None | Percent Complete: | 0% |
Open/Closed: | Open | Effort: | 0.00 |
No files currently attached
Depends on the following items: None found
Items that depend on this one: None found
Carbon-Copy List
There are 0 votes so far. Votes easily highlight which items people would like to see resolved in priority, independently of the priority of the item set by tracker managers.
No changes have been made to this item
Powered by Savane 3.13-f8d8.
Corresponding source code
There is a size (around 1e5 connections) at which continuous networks (requiring to process every connection every time) become intolerably inefficient.
Spiking networks (in which we keep a list of only those nodes which have had at least one 'spike' in their input, and don't touch any connections originating at other nodes) are necessary for efficient handling of large problems.
It is likely that this incremental interleaving of task with task-tracking will mean we lose most of the benefit of GPU or OMP implementations, so spiking will need to be saving at least 95% of the connection processing in order for the optimization to be worthwhile. Therefore we will probably need two implementations of spiking!feedforward (one multiprocessing and one task-tracking) and analysis/tracking code to distinguish which which is likely to be faster.