Autonomous navigation using spiking neuromorphic computers

Patent metadata

Field	Value
Publication number	US-10846590-B2
Application number	US-201615385299-A
Country	US
Kind code	B2
Filing date	Dec 20, 2016
Priority date	Dec 20, 2016
Publication date	Nov 24, 2020
Grant date	Nov 24, 2020

Abstract

A spike timing dependent plasticity (STDP) rule is applied in a spiking neural network (SNN) that includes artificial synapses bi-directionally connecting artificial neurons in the SNN to model locations within a physical environment. A first neuron is activated to cause a spike wave to propagate from the first neuron to other neurons in the SNN. Propagation of the spike wave causes synaptic weights of a subset of the synapses to be increased based on the STDP rule. A second neuron is activated after propagation of the spike wave to cause a spike chain to propagate along a path from the second neuron to the first neuron, based on the changes to the synaptic weights. A physical path is determined from the second to the first neuron based on the spike chain, and a signal may be sent to a controller of an autonomous device to cause the autonomous to navigate the physical path.

First claim

What is claimed is:

1. At least one non-transitory machine accessible storage medium having instructions stored thereon, wherein the instructions when executed on a machine, cause the machine to:
define a spike timing dependent plasticity (STDP) rule to be applied in a spiking neural network (SNN), wherein the SNN comprises a plurality of artificial synapses to bi-directionally connect neurons in a plurality of artificial neurons to model a plurality of locations within a physical environment:
send a first signal to activate a first neuron in the plurality of artificial neurons, wherein the first signal is to designate a first location in the plurality of locations as a destination location, activating the first neuron causes a spike wave to propagate from the first neuron to other neurons in the plurality of artificial neurons, propagation of the spike wave causes synaptic weights of a subset of synapses in the plurality of synapses to be increased based on the STDP rule, and the subset of synapses comprises synapses to transmit spikes in a direction opposed to the propagation of the spike wave;
send a neuromodulatory signal after propagation of the spike wave to cause an increase in a spiking threshold potential parameter of each of the plurality of artificial neurons;
send a second signal to activate a second neuron in the plurality of artificial neurons, wherein the second signal is to be sent after propagation of the spike wave and is to designate a second location in the plurality of locations as a starting location, and activation of the second neuron causes a spike chain to propagate along a path from the second neuron to the first neuron; and
send a third signal to a controller of an autonomous device to cause the autonomous device to navigate a physical path from the starting location to the destination location based on the path.

2. The storage medium of claim 1 , wherein each of the plurality of artificial neurons corresponds to a respective one of the plurality of locations, the first neuron is defined to correspond to the first location, and the second neuron is defined to correspond to the second location.

3. The storage medium of claim 2 , wherein the spike chain propagates to a subset of the plurality of artificial neurons on a portion of the subset of synapses, and the path comprises a path over locations represented by the subset of neurons.

4. The storage medium of claim 1 , wherein the SNN is defined to model a lattice network corresponding coordinates in the physical environment.

5. The storage medium of claim 4 , wherein the lattice network comprises one of a two-dimensional lattice to model a two-dimensional physical space or a three-dimensional lattice to model a three-dimensional physical space.

6. The storage medium of claim 1 , wherein the neuromodulatory signal is to increase each spiking threshold potential parameter to a value that ensures that a single presynaptic spike trigger a postsynaptic spike only if the presynaptic spike was sent by one of the synapses in the subset of synapses.

7. The storage medium of claim 1 , wherein the instructions, when executed, further cause a machine to:
send a fourth signal after the first and second signals, wherein the fourth signal is to activate a third neuron in the plurality of artificial neurons, the fourth signal is to designate a third location corresponding to a third location as a different starting location, and activation of the third neuron causes another spike chain to propagate along a particular path from the third neuron to the first neuron; and
determine a path from the third location to the destination location based on the particular path.

8. A method comprising:
defining a spike timing dependent plasticity (STDP) rule to be applied in a spiking neural network (SNN), wherein the SNN comprises a plurality of artificial synapses to bi-directionally connect neurons in a plurality of artificial neurons to model a plurality of locations within a physical environment;
sending a first signal to activate a first neuron in the plurality of artificial neurons, wherein the first signal is to designate a first location in the plurality of locations as a destination location, activating the first neuron causes a spike wave to propagate from the first neuron to other neurons in the plurality of artificial neurons, propagation of the spike wave causes synaptic weights of a subset of synapses in the plurality of synapses to be increased based on the STDP rule, and the subset of synapses comprises synapses to transmit spikes in a direction opposed to the propagation of the spike wave;
sending a neuromodulatory signal after propagation of the spike wave to cause an increase in a spiking threshold potential parameter of each of the plurality of artificial neurons;
sending a second signal to activate a second neuron in the plurality of artificial neurons, wherein the second signal is to be sent after propagation of the spike wave and is to designate a second location in the plurality of locations as a starting location, and activation of the second neuron causes a spike chain to propagate along a path from the second neuron to the first neuron; and
sending a third signal to a controller of an autonomous device to cause the autonomous device to navigate a physical path from the starting location to the destination location based on the path.

9. An apparatus comprising:
a neuromorphic computing device comprising:
a routing fabric;
a plurality of neuromorphic cores, wherein at least a subset of the plurality of neuromorphic cores are to respectively implement one or more of a plurality of artificial neurons in a spiking neural network (SNN) to model a physical environment, and each neuron in the SNN models a respective one of a plurality of locations within the physical environment,

wherein the neuromorphic computing device is to:
route a first signal to activate a first neuron in the plurality of artificial neurons, wherein the first neuron corresponds to a first one of the plurality of locations;
generate a first set of spikes to be sent by the first neuron in response to activation of the first neuron, wherein the first set of spikes causes a spike wave in the SNN;
propagate the spike wave to other neurons in the plurality of artificial neurons on artificial synapses in the SNN, wherein the artificial synapses are implemented using the routing fabric;
adjust synaptic weights of a subset of the synapses based on propagation of the spike wave and a spike timing dependent plasticity (STDP) rule, and the subset of synapses comprises synapses to transmit spikes in a direction opposed to the propagation of the spike wave;
send a neuromodulatory signal after propagation of the spike wave to cause an increase in a spiking threshold potential parameter of each of the plurality of artificial neurons;
route a second signal to activate a second neuron in the plurality of artificial neurons, wherein the second neuron corresponds to a second one of the plurality of locations;
generate a second set of spikes to be sent by the second neuron in response activation of the second neuron;
propagate a spike chain across a portion of the subset of synapses, wherein the spike chain is initiated by the second set of spikes, and the spike chain is to propagate along a path defined by the portion of the subset of synapses from the second neuron to the first neuron in the SNN.

10. The apparatus of claim 9 , wherein the neuromorphic computing device further comprises logic to generate data to describe the path, wherein the path comprises a subset of the neurons and corresponds to a path in the physical environment comprising locations modeled by the subset of

Assignees

• Intel Corp

Inventors

• Imam Nabil
• Srinivasa Narayan

Classifications

• G06N3/065

  Analogue means · CPC title

• G06N3/082

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

• G06N3/0495

  Quantised networks; Sparse networks; Compressed networks · CPC title

• G06N3/088

  Non-supervised learning, e.g. competitive learning · CPC title

• G06N3/063

  using electronic means · CPC title