SirStone/SirRoboGarage
1
0
Fork 0
Code Issues 28 Pull Requests Actions Projects Releases Wiki Activity

Research PC-SNN as backup learning algorithm #157

New Issue
Closed
opened 2026-09-13 19:19:54 +02:00 by SirStone · 1 comment
SirStone commented 2026-09-13 19:19:54 +02:00
Owner
Copy Link
Copy Source

Question

Is PC-SNN (Predictive Coding SNN, arXiv:2211.15386) a viable backup if SuperSpike doesn't converge for our angle regression task? Specifically:

  • How does predictive coding work in an SNN context?
  • Has it been tested on angle/continuous regression tasks?
  • How does implementation complexity compare to SuperSpike?
  • What are the key hyperparameters?

Context: This is the fallback if SuperSpike doesn't work for SNNBot's 36→12→2 topology aiming at a stationary target.

Blocked by #156 (SuperSpike research should be evaluated first).

Related to wayfinder map #147.

## Question Is PC-SNN (Predictive Coding SNN, arXiv:2211.15386) a viable backup if SuperSpike doesn't converge for our angle regression task? Specifically: - How does predictive coding work in an SNN context? - Has it been tested on angle/continuous regression tasks? - How does implementation complexity compare to SuperSpike? - What are the key hyperparameters? Context: This is the fallback if SuperSpike doesn't work for SNNBot's 36→12→2 topology aiming at a stationary target. Blocked by [#156](https://git.fossellini.top/SirStone/SirRoboGarage/issues/156) (SuperSpike research should be evaluated first). Related to wayfinder map [#147](https://git.fossellini.top/SirStone/SirRoboGarage/issues/147).
SirStone added the wayfinder:research label 2026-09-13 19:19:54 +02:00
SirStone commented 2026-09-13 19:23:02 +02:00
Author
Owner
Copy Link
Copy Source

Resolution

PC-SNN is viable but too costly as a backup. It requires:

  • Full rewrite to time-to-first-spike (TTFS) coding — incompatible with current rate/population coding
  • Per-layer error nodes + 10 iterative inference passes per sample
  • More code and tuning surface than SuperSpike

SuperSpike (#156) is the clear winner — same locality guarantees, compatible with existing coding scheme, simpler implementation.

Full research: SNNBot_garage/research/pc-snn.md on branch research/pc-snn

## Resolution PC-SNN is viable but too costly as a backup. It requires: - Full rewrite to time-to-first-spike (TTFS) coding — incompatible with current rate/population coding - Per-layer error nodes + 10 iterative inference passes per sample - More code and tuning surface than SuperSpike SuperSpike (#156) is the clear winner — same locality guarantees, compatible with existing coding scheme, simpler implementation. Full research: `SNNBot_garage/research/pc-snn.md` on branch `research/pc-snn`
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
ready-for-agent
wayfinder:grilling
Wayfinder design question to grill
 wayfinder:map
Wayfinder map issue
 wayfinder:prototype
wayfinder:research
Wayfinder research question
 wayfinder:task
Wayfinder implementation task
No Label wayfinder:research
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
SirStone
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: SirStone/SirRoboGarage#157
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?