Efficient generation of stochastic spike patterns in core-based neuromorphic systems

Patent metadata

Field	Value
Publication number	US-11574183-B2
Application number	US-201916539689-A
Country	US
Kind code	B2
Filing date	Aug 13, 2019
Priority date	Jan 13, 2016
Publication date	Feb 7, 2023
Grant date	Feb 7, 2023

Abstract

Weighted population code in neuromorphic systems is provided. According to an embodiment, a plurality of input values is received. For each of the plurality of values, a plurality of spikes is generated. Each of the plurality of spikes has an associated weight. A consumption time is determined for each of the plurality of spikes. Each of the plurality of spikes is sent for consumption at its consumption time.

First claim

What is claimed is:

1. A method comprising:
receiving a plurality of input values;
for each of the plurality of values, generating a plurality of spikes, each of the plurality of spikes having an associated weight, wherein each of the associated weights is selected from a predetermined set of weights, wherein the set of weights is determined based on a range of the plurality of input values, wherein the set of weights is determined to minimize a cardinality of the plurality of spikes for those of the plurality of values within a predetermined range;
determining a consumption time for each of the plurality of spikes;
sending each of the plurality of spikes for consumption at its consumption time.

2. The method of claim 1 , wherein the consumption time for each of the plurality of spikes is selected from a predetermined set of consumption times.

3. The method of claim 1 , wherein sending each of the plurality of spikes comprises sending a delay instruction corresponding to its consumption time.

4. The method of claim 1 , wherein the consumption time of the plurality of spikes is within a predetermined setup period.

5. A method comprising:
receiving at a first neuromorphic system a first plurality of spikes, the first plurality of spikes encoding a first value according to a first encoding, wherein the first encoding associates a weight with each of the first plurality of spikes, wherein the weights of the first plurality of spikes are determined to minimize a cardinality of the first plurality of spikes for those of a plurality of values within a predetermined range;
generating from the first plurality of spikes a second plurality of spikes, the second plurality of spikes encoding the first value according to a second encoding;
providing the second plurality of spikes to a second neuromorphic system.

6. The method of claim 5 , further comprising:
receiving at the first neuromorphic system a plurality of delays, each of the plurality of delays associated with one of the first plurality of spikes;
delaying each of the first plurality of spikes by its associated delay.

7. The method of claim 5 , wherein receiving the first plurality of spikes comprises:
receiving a first subset of the first plurality of spikes during a first time window;
receiving a second subset of the first plurality of spikes during a second time window.

8. The method of claim 5 , wherein generating the second plurality of spikes comprises:
applying the weights of the first encoding to the first plurality of spikes.

9. The method of claim 5 , wherein each of the first plurality of spikes is received in one of a predetermined number of time windows.

10. The method of claim 5 , wherein each of the plurality of spikes represents a discrete value.

11. The method of claim 5 , wherein generating the second plurality of spikes comprises:
determining a probability from the weights of the first encoding and the first plurality of spikes.

12. The method of claim 5 , further comprising:
providing output to the second neuromorphic system, and concurrently therewith, receiving at the first neuromorphic system a third plurality of spikes, the third plurality of spikes encoding a second value according to the first encoding.

13. The method of claim 5 , further comprising:
receiving at the first neuromorphic system a third plurality of spikes, the third plurality of spikes encoding the difference between a second value and the first value according to the first encoding;
generating from the third plurality of spikes a fourth plurality of spikes, the fourth plurality of spikes encoding the second value according to the second encoding;
providing the fourth plurality of spikes to the second neuromorphic system.

14. The method of claim 5 , wherein the second neuromorphic system comprises a classifier.

15. A method comprising:
receiving at a neuromorphic system a first plurality of spikes, the first plurality of spikes encoding a first value according to a first encoding, wherein the first encoding associates a weight with each of the first plurality of spikes, wherein the weights of the first plurality of spikes are determined to minimize a cardinality of the first plurality of spikes for those of a plurality of values within a predetermined range;
generating from the first plurality of spikes a plurality of sets of spikes, each of the plurality of sets of spikes encoding the first value according to a second encoding;
providing each of the plurality of sets of spikes to one of a plurality of classifiers in the neuromorphic system.

16. A system comprising:
a neuromorphic network comprising a plurality of input axons, and a plurality of stochastic neurons connected thereto;
an encoder operatively coupled to the plurality of input axons, the encoder configured to send to the input axons a first plurality of spikes, the first plurality of spikes encoding a first value according to a first encoding, wherein the first encoding associates a weight with each of the first plurality of spikes, wherein the weights of the first plurality of spikes are determined to minimize a cardinality of the first plurality of spikes for those of a plurality of values within a predetermined range;
the neuromorphic network configured to generate from the first plurality of spikes a second plurality of spikes, the second plurality of spikes encoding the first value according to a second encoding.

17. The method of claim 15 , further comprising:
receiving at the first neuromorphic system a plurality of delays, each of the plurality of delays associated with one of the first plurality of spikes;
delaying each of the first plurality of spikes by its associated delay.

18. The method of claim 15 , wherein receiving the first plurality of spikes comprises:
receiving a first subset of the first plurality of spikes during a first time window;
receiving a second subset of the first plurality of spikes during a second time window.

19. The method of claim 15 , wherein generating the plurality of sets of spikes comprises:
applying the weights of the first encoding to the first plurality of spikes.

20. The method of claim 15 , wherein generating the plurality of sets of spikes comprises:
determining a probability from the weights of the first encoding and the first plurality of spikes.

Assignees

• IBM

Inventors

• Amir Arnon
• Jimeno Yepes Antonio J
• Tang Jianbin

Classifications

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G06N3/065 Primary

  Analogue means · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/063

  using electronic means · CPC title

• G06N3/049 Primary

  Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs · CPC title