Method and apparatus for modeling neural resource based synaptic placticity
Certain aspects of the present disclosure support a method of designing the resource model in hardware (or software) for learning spiking neural networks. The present disclosure comprises accounting for resources in a different domain (e.g., negative log lack-of-resources instead of availability of resources), modulating weight changes for multiple spike events upon a single trigger, and strategically advancing or retarding the resource replenishment or decay (respectively) to overcome the limitation of single event-based triggering.
1. A method of learning in a spiking neural network, comprising:
modeling resources associated with synapses of the spiking neural network using a defined domain;
modulating changes of a weight associated with one of the synapses for multiple spike events upon a single trigger related to that synapse; and
updating, in the defined domain, one of the resources associated with that synapse based on the multiple spike events upon the single trigger.
2. The method of claim 1 , wherein the defined domain is represented with negative log lack-of resource availability.
3. The method of claim 1 , wherein the defined domain is represented with lack of resource availability or with log of resource availability.
4. The method of claim 1 , wherein modulating the changes of the weight comprises:
modulating a long-term depression (LTD) change of the weight; and
modulating a long-term potentiation (LTP) change of the weight.
5. The method of claim 4 , wherein updating that one resource comprises:
recovering that one resource prior to modulating the LTD change of weight; and
recovering that one resource prior to modulating the LTP change of weight.
6. The method of claim 4 , wherein updating that one resource comprises:
eroding that one resource following modulating the LTD change of weight; and
eroding that one resource following modulating the LTP change of weight.
7. The method of claim 4 , further comprising:
saving a state of that one resource in the defined domain after performing both the LTD change of weight and the LTP change of weight.
8. The method of claim 1 , further comprising:
projecting that one resource back in time to a replay time of a pre-synaptic spike of the multiple spike events.
9. The method of claim 8 , wherein projecting that one resource comprises:
projecting decay of that one resource back in time.
10. An apparatus for learning in a spiking neural network, comprising:
a first circuit configured to model resources associated with synapses of the spiking neural network using a defined domain;
a second circuit configured to modulate changes of a weight associated with one of the synapses for multiple spike events upon a single trigger related to that synapse; and
a third circuit configured to update, in the defined domain, one of the resources associated with that synapse based on the multiple spike events upon the single trigger.
11. The apparatus of claim 10 , wherein the defined domain is represented with negative log lack-of resource availability.
12. The apparatus of claim 10 , wherein the defined domain is represented with lack of resource availability or with log of resource availability.
13. The apparatus of claim 10 , wherein the second circuit is configured to modulate changes of the weight by:
modulating a long-term depression (LTD) change of the weight; and
modulating a long-term potentiation (LTP) change of the weight.
14. The apparatus of claim 13 , wherein the third circuit is configured to update that one resource by:
recovering that one resource prior to modulating the LTD change of weight; and
recovering that one resource prior to modulating the LTP change of weight.
15. The apparatus of claim 13 , wherein the third circuit is configured to update that one resource by:
eroding that one resource following modulating the LTD change of weight; and
eroding that one resource following modulating the LTP change of weight.
16. The apparatus of claim 13 , further comprising:
a fourth circuit configured to save a state of that one resource in the defined domain after performing both the LTD change of weight and the LTP change of weight.
17. The apparatus of claim 10 , further comprising:
a fourth circuit configured to project that one resource back in time to a replay time of a pre-synaptic spike of the multiple spike events.
18. The apparatus of claim 17 , wherein projecting that one resource comprises:
projecting decay of that one resource back in time.
19. An apparatus for learning in a spiking neural network, comprising:
means for modeling resources associated with synapses of the spiking neural network using a defined domain;
means for modulating changes of a weight associated with one of the synapses for multiple spike events upon a single trigger related to that synapse; and
means for updating, in the defined domain, one of the resources associated with that synapse based on the multiple spike events upon the single trigger.
20. The apparatus of claim 19 , wherein the defined domain is represented with negative log lack-of resource availability.
21. The apparatus of claim 19 , wherein the defined domain is represented with lack of resource availability or with log of resource availability.
22. The apparatus of claim 19 , further comprising:
means for modulating a long-term depression (LTD) change of the weight; and
means for modulating a long-term potentiation (LTP) change of the weight.
23. The apparatus of claim 22 , further comprising:
means for recovering that one resource prior to modulating the LTD change of weight; and
means for recovering that one resource prior to modulating the LTP change of weight.
24. The apparatus of claim 22 , further comprising:
means for eroding that one resource following modulating the LTD change of weight; and
means for eroding that one resource following modulating the LTP change of weight.
25. The apparatus of claim 22 , further comprising:
means for saving a state of that one resource in the defined domain after performing both the LTD change of weight and the LTP change of weight.
26. The apparatus of claim 19 , further comprising:
means for projecting that one resource back in time to a replay time of a pre-synaptic spike of the multiple spike events.
27. The method of claim 26 , wherein the means for projecting that one resource comprises:
means for projecting decay of that one resource back in time.
28. A computer program product for learning in a spiking neural network, comprising a computer-readable medium comprising code for:
modeling resources associated with synapses of the spiking neural network using a defined domain;
modulating changes of a weight associated with one of the synapses for multiple spike events upon a single trigger related to that synapse; and
updating, in the defined domain, one of the resources associated with that synapse based on the multiple spike events upon the single trigger.
29. The computer program product of claim 28 , wherein the defined domain is represented with negative log lack-of resource availability.
30. The computer program product of claim 28 , wherein the defined domain is represented with lack of resource availability or with log of resource availability.
31. The computer program product of claim 28 , wherein the computer-readable medium further comprising code for:
modulating a long-term depression (LTD) change of the weight; and
modulating a long-term potentiation (LTP) change of the weight.
32. The computer program product of claim 31 , wherein the computer-readable medium further comprising code for:
recovering that one resource prior to modulating the LTD change of weight; and
recovering that one resource prior to modulating the LTP change of weight.
33. The computer program product of claim 31 , wherein the computer-readable medium further comprising code for:
eroding that one resource following modulating the LTD change of weight; and
eroding that one resource following modulating the LTP change of weight.
34. The computer program product of claim 31 , wherein the computer-readable medium further comprising code for:
saving a state of that one resource in the defined domain after performing both the LTD change of weight and the LTP change of weight.
35. The computer program product of claim 28 , wherein the computer-readable medium further comprising code for:
projecting that one resource back in time to a replay time of a pre-synaptic spike of the multiple spike events.
36. The computer program product of claim 35 , wherein the computer-readable medium further comprising code for:
projecting decay of that one resource back in time.