Plastic synapse management

Patent metadata

Field	Value
Publication number	US-9542645-B2
Application number	US-201414228078-A
Country	US
Kind code	B2
Filing date	Mar 27, 2014
Priority date	Mar 27, 2014
Publication date	Jan 10, 2017
Grant date	Jan 10, 2017

Abstract

A method for managing synapse plasticity in a neural network includes converting a first set of synapses from a plastic synapse type to a fixed synapse type. The method may also include converting a second set of synapses from the fixed synapse type to the plastic synapse type.

First claim

What is claimed is:

1. A method for managing synapse plasticity in an artificial neural network, comprising
determining whether a number of plastic synapse types defined for the artificial neural network is greater than a hardware limit; and
converting at least one of: a first set of synapses from a plastic synapse type to a fixed synapse type when the first set of synapses is designated as unused, a second set of synapses from the fixed synapse type to the plastic synapse type when the second set of synapses is designated as used, or a combination thereof, when the number of defined plastic synapse types is greater than the hardware limit, such that a number of active plastic synapse types is less than or equal to the hardware limit.

2. The method of claim 1 , in which designating a set of synapses as used or unused is based at least in part on one of: a user input, power management, computation time management, a minimum number of synapse instances, a synapse training interval, a synapse type, an overall training need of the artificial neural network, or a combination thereof.

3. The method of claim 1 , in which the second set of synapses is designated to be trained in a neural model.

4. The method of claim 1 , in which the number of plastic synapse types is pre-defined.

5. The method of claim 1 , in which the converting is controlled by a resource manager.

6. The method of claim 1 , further comprising changing properties of the first set of synapses and the second set of synapses based on outputs of the artificial neural network and/or performance of the artificial neural network.

7. The method of claim 1 , further comprising performing plasticity related computations in a neuron when all outgoing and/or incoming synapses of the neuron are non-plastic synapses.

8. An apparatus for managing synapse plasticity in an artificial neural network, comprising:
a memory unit; and
at least one processor coupled to the memory unit, the at least one processor being configured:
to determine whether a number of plastic synapse types defined for the artificial neural network is greater than a hardware limit; and
to convert at least one of: a first set of synapses from a plastic synapse type to a fixed synapse type when the first set of synapses is designated as unused, a second set of synapses from the fixed synapse type to the plastic synapse type when the second set of synapses is designated as used, or a combination thereof, when the number of defined plastic synapse types is greater than the hardware limit, such that a number of active plastic synapse types is less than or equal to the hardware limit.

9. The apparatus of claim 8 , in which the at least one processor is further configured to designate a set of synapses as used or unused based at least in part on one or more of a user input, power management, computation time management, a minimum number of synapse instances, a synapse training interval, a synapse type, an overall training need of the artificial neural network, or a combination thereof.

10. The apparatus of claim 8 , in which the second set of synapses is designated to be trained in a neural model.

11. The apparatus of claim 8 , in which the number of plastic synapse types is pre-defined.

12. The apparatus of claim 8 , in which the converting is controlled by a resource manager.

13. The apparatus of claim 8 , in which the at least one processor is further configured to change properties of the first set of synapses and the second set of synapses based on at least one of outputs of the artificial neural network and/or performance of the artificial neural network.

14. The apparatus of claim 8 , in which the at least one processor is further configured to perform plasticity related computations in a neuron based on whether all outgoing and/or incoming synapses of the neuron are fixed synapses.

15. An apparatus for managing synapse plasticity in an artificial neural network, comprising
means for determining whether a number of plastic synapse types defined for the artificial neural network is greater than a hardware limit of the artificial neural network; and
means for converting at least one of: a first set of synapses from a plastic synapse type to a fixed synapse type when the first set of synapses is designated as unused, a second set of synapses from the fixed synapse type to the plastic synapse type when the second set of synapses is designated as used, or a combination thereof, when the number of defined plastic synapse types is greater than the hardware limit, such that a number of active plastic synapse types is less than or equal to the hardware limit.

16. The apparatus of claim 15 , in which designating a set of synapses as used or unused is based at least in part on one or more of a user input, power management, computation time management, a minimum number of synapse instances, a synapse training interval, a synapse type, an overall training need of the artificial neural network, or a combination thereof.

17. The apparatus of claim 16 , in which the second set of synapses is designated to be trained in a neural model.

18. A non-transitory computer-readable medium having program code recorded thereon for managing synapse plasticity in an artificial neural network, the program code comprising:
program code to determine whether a number of plastic synapse types defined for the artificial neural network is greater than a hardware limit; and
program code to convert at least one of: a first set of synapses from a plastic synapse type to a fixed synapse type when the first set of synapses is designated as unused, a second set of synapses from the fixed synapse type to the plastic synapse type when the second set of synapses is designated as used, or a combination thereof, when the number of defined plastic synapse types is greater than the hardware limit, such that a number of active plastic synapse types is less than or equal to the hardware limit.

19. The non-transitory computer-readable medium of claim 18 , in which the program code further comprises code to designate a set of synapses as used or unused based at least in part on one or more of a user input, power management, computation time management, a minimum number of synapse instances, a synapse training interval, a synapse type, an overall training need of the artificial neural network, or a combination thereof.

20. The non-transitory computer-readable medium of claim 18 , in which the second set of synapses is designated to be trained in a neural model.

Assignees

• Qualcomm Inc

Inventors

• Gupta Vikram
• Gibson Sarah Paige
• Levin Jeffrey Alexander
• Patwardhan Ravindra Manohar
• Chakraborty Avijit
• Constable William Howard
• Bell Ii William Richard

Classifications

• G06N3/065 Primary

  Analogue means · CPC title

• G06N3/049 Primary

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

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/082

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title

• G06N3/08 Primary

  Learning methods · CPC title