Apparatus and methods for rate-modulated plasticity in a neuron network

Patent metadata

Field	Value
Publication number	US-9460385-B2
Application number	US-201414466917-A
Country	US
Kind code	B2
Filing date	Aug 22, 2014
Priority date	Feb 22, 2013
Publication date	Oct 4, 2016
Grant date	Oct 4, 2016

Abstract

Apparatus and methods for activity based plasticity in a spiking neuron network adapted to process sensory input. In one approach, the plasticity mechanism of a connection may comprise a causal potentiation portion and an anti-causal portion. The anti-causal portion, corresponding to the input into a neuron occurring after the neuron response, may be configured based on the prior activity of the neuron. When the neuron is in low activity state, the connection, when active, may be potentiated by a base amount. When the neuron activity increases due to another input, the efficacy of the connection, if active, may be reduced proportionally to the neuron activity. Such functionality may enable the network to maintain strong, albeit inactive, connections available for use for extended intervals.

First claim

What is claimed:

1. An apparatus, comprising:
a memory; and
at least one processor coupled to the memory, the at least one processor configured, based on a response by a neuron:
to increase an efficacy of a connection configured to provide input to the neuron prior to the response; and
to adjust the efficacy of the connection configured to provide input to the neuron subsequent to the response;

wherein:
the adjustment of the efficacy is determined based on a rate of the response;
the response comprises one or more spikes generated by the neuron based on the input;
the rate of the response comprises an average rate determined based on at least a number of the one or more spikes occurring within an interval preceding the response;
the increasing of the efficacy is characterized by a first rule; and
the adjustment of the efficacy comprises increasing the efficacy based on a second rule, a magnitude of the increase being determined based at least on the average rate being below a threshold.

2. The apparatus of claim 1 wherein:
the input is characterized by a refresh period of 40 ms; and
the rate of the response is based on an average rate of responses by the neuron determined within a time window selected from the range between 1 s and 2000 s, inclusive.

3. The apparatus of claim 1 wherein the adjustment of the efficacy further comprises a decrease in the efficacy, a magnitude of the decrease being based on the average rate being above the threshold.

4. The apparatus of claim 3 wherein the magnitude of the decrease and the magnitude of the increase are each independent of a time interval between the response and the input.

5. The apparatus of claim 4 wherein the magnitude of the increase is configured to be smaller than the magnitude of the decrease, and the magnitude of the increase is characterized by the first rule.

6. The apparatus of claim 4 , wherein:
the input is characterized by a refresh rate of 25 Hz;
the threshold is selected from the range between 0.05 Hz and 0.2 Hz; and
a ratio of the magnitude of the increase based on the second rule to the magnitude of the decrease is selected from the range between 0.05 and 0.2 inclusive.

7. The apparatus of claim 1 wherein the adjustment is based at least in part on a time interval between the response and another response preceding the response.

8. The apparatus of claim 1 wherein:
the efficacy increase comprises increasing the efficacy based on a delay between the response and the input being below the threshold; and
the efficacy is maintained based on the delay between the response and the input being above the threshold.

9. The apparatus of claim 8 , wherein:
the input is characterized by a refresh rate;
the refresh rate comprises 40 ms; and
the threshold is selected from the range between 500 ms and 1000 ms, inclusive.

10. The apparatus of claim 1 wherein:
the input is characterized by a refresh rate of 40 ms; and
the efficacy increase comprises increasing the efficacy when delay between the response and the input is below the threshold,
the threshold is selected from the range between 20 ms and 200 ms, inclusive.

11. A sensory processing spiking neuron network apparatus, comprising:
a connection configured to provide an input to a neuron configured to generate a response based on the input;
wherein the connection is further configured to be:
potentiated when the input is within an interval from the response;
adjusted when the input occurs subsequent to the response, the adjustment being determined based on activity of the neuron prior to the response.

12. The apparatus of claim 11 , wherein:
the connection is characterized by a connection efficacy configured to delay or advance a response generation by the neuron;
the potentiation comprises increase of the efficacy configured to advance generation of another response subsequent to the response;
if the response occurs after a second time interval after the input, the efficacy is maintained at a level corresponding to the efficacy prior to the response; and
the second time interval comprises at least a portion of the interval.

13. A non-transitory computer-readable medium having encoded thereon program code, the program code executed by a processor and comprising:
program code to update a connection configured to provide stimulus to an artificial neuron
program code to potentiate the connection if the stimulus precedes a response generated by the artificial neuron;
program code to potentiate the connection if the response precedes the stimulus and a neuron activity is below a threshold level; and ;and
program code to depress the connection when the response precedes the stimulus and the neuron activity is above the threshold level.

14. The non-transitory computer-readable medium of claim 13 , wherein:
the depression of the connection is configured to delay another response by the artificial neuron subsequent to the response; and
the potentiation of the connection is configured to advance the another response.

15. The non-transitory computer-readable medium of claim 14 , wherein:
the connection is characterized by a latency configured to delay provision of the stimulus to the artificial neuron;
the connection potentiation comprises a reduction in the latency; and
the connection depression comprises an increase in the latency.

16. The non-transitory computer-readable medium of claim 15 , wherein:
the connection is characterized by a weight configured to affect generation of the response by the artificial neuron such that a larger weight corresponds to a greater probability of the response being generated;
the connection potentiation comprises an increase in the weight; and
the connection depression comprises a reduction in the weight.

17. A method of managing a connection in a neuron network based on at least one signal from a neuron, the method comprising:
receiving at least one input via the connection;
sending the at least one signal at a time proximate to the received at least one input;
if the at least one signal is sent prior to the reception of the input, demoting the connection; and
if the at least one signal is sent after the reception of the input, promoting the connection.

18. The method of claim 17 , wherein:
the demotion of the connection decreases a probability associated with signal generation at the neuron; and
the promotion of the connection increases a probability associated with signal generation at the neuron.

19. The method of claim 17 , wherein:
the demotion of the connection is configured to delay signal generation at the neuron; and
the promotion of the connection is configured to advance signal generation at the neuron.

20. The method of claim 17 , wherein the time proximate to the received at least one input comprises an interval during which the at least one input is received.

21. The method of claim 20 , wherein a neuron response is governed by a minimum feature time scale, the interval being greater than the minimum feature time scale.

22. An apparatus comprising
means for increasing an efficacy of a connection configured to provide input to a neuron prior to a response by the neuron; and
means for adjusting the efficacy of the connection configured to provide input to the neuron subsequent to the response;
wherein:
the adjustment of the efficacy is determined based on a rate of the response;
the response comprises one or more spikes generated by

Assignees

• Qualcomm Technologies Inc

Inventors

• Piekniewski Filip
• Richert Micah
• Izhikevich Eugene
• Chan Victor Hokkiu
• Levin Jeffrey Alexander

Classifications

• G06N3/049 Primary

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

• G06N3/0499

  Feedforward networks · CPC title

• G06N99/005

  Physics · mapped topic

• G06N3/10

  Interfaces, programming languages or software development kits, e.g. for simulating neural networks · CPC title

• G06N3/02

  Neural networks · CPC title