Neuronal network topology for computing conditional probabilities

Patent metadata

Field	Value
Publication number	US-10748063-B2
Application number	US-201916294815-A
Country	US
Kind code	B2
Filing date	Mar 6, 2019
Priority date	Apr 17, 2018
Publication date	Aug 18, 2020
Grant date	Aug 18, 2020

Abstract

Described is a system for estimating conditional probabilities for operation of a mobile device. Input data streams from first and second mobile device sensors are input into a neuronal network, where the first and second input data streams are converted into variable spiking rates of first and second neurons. The system learns a conditional probability between the first and second input data streams. A synaptic weight of interest between the first and second neurons converges to a fixed-point value, where the fixed-point value corresponds to the conditional probability. Based on the conditional probability and a new input data stream, a probability of an event is estimated. Based on the probability of the event, the system causes the mobile device to perform a mobile device operation.

First claim

What is claimed is:

1. A system for estimating conditional probabilities for operation of a mobile device, the system comprising:
one or more processors and a non-transitory computer-readable medium having executable instructions encoded thereon such that when executed, the one or more processors perform operations of:
inputting a first input data stream obtained from a first mobile device sensor and a second input data stream obtained from a second mobile device sensor into a neuronal network comprising a plurality of neurons, wherein the first and second input data streams are converted into variable spiking rates of a first neuron and a second neuron;
learning a conditional probability between the first input data stream and the second input data stream,
wherein a synaptic weight of interest between the first neuron and the second neuron converges to a fixed-point value corresponding to the conditional probability;
based on the conditional probability and a new input data stream, estimating a probability of an event; and
based on the probability of the event, causing the mobile device to perform a mobile device operation.

2. The system as set forth in claim 1 , wherein the mobile device operation is a collision avoidance maneuver.

3. The system as set forth in claim 1 , wherein the mobile device operation is generation of an alert providing instructions to a mobile device operator.

4. The system as set forth in claim 1 , wherein the synaptic weight of interest is updated each time either of the first and second neurons spike according to spike-timing dependent plasticity (STDP), causing a corresponding update in the conditional probability such that the conditional probability is adapted in real-time.

5. The system as set forth in claim 4 , wherein all synaptic connections between neurons in the neuronal network have a predetermined delay, and wherein all synaptic weights besides the synaptic weight of interest are fixed at a value such that only the synaptic weight of interest is updated according to STDP.

6. The system as set forth in claim 5 , wherein an increment or decrement in the synaptic weight of interest due to spikes in the first neuron causing spikes in the second neuron is set to a constant value that is a multiplier on a change in the synaptic weight of interest.

7. The system as set forth in claim 1 , wherein tonic and phasic inputs are used to stabilize the synaptic weight of interest.

8. A computer implemented method for estimating conditional probabilities for operation of a mobile device, the method comprising an act of:
causing one or more processers to execute instructions encoded on a non-transitory computer-readable medium, such that upon execution, the one or more processors perform operations of:
inputting a first input data stream obtained from a first mobile device sensor and a second input data stream obtained from a second mobile device sensor into a neuronal network comprising a plurality of neurons, wherein the first and second input data streams are converted into variable spiking rates of a first neuron and a second neuron;
learning a conditional probability between the first input data stream and the second input data stream,
wherein a synaptic weight of interest between the first neuron and the second neuron converges to a fixed-point value corresponding to the conditional probability;
based on the conditional probability and a new input data stream, estimating a probability of an event; and
based on the probability of the event, causing the mobile device to perform a mobile device operation.

9. The method as set forth in claim 8 , wherein the mobile device operation is a collision avoidance maneuver.

10. The method as set forth in claim 8 , wherein the mobile device operation is generation of an alert providing instructions to a mobile device operator.

11. The method as set forth in claim 8 , wherein the synaptic weight of interest is updated each time either of the first and second neurons spike according to spike-timing dependent plasticity (STDP), causing a corresponding update in the conditional probability such that the conditional probability is adapted in real-time.

12. The method as set forth in claim 11 , wherein all synaptic connections between neurons in the neuronal network have a predetermined delay, and wherein all synaptic weights besides the synaptic weight of interest are fixed at a value such that only the synaptic weight of interest is updated according to STDP.

13. The method as set forth in claim 12 , wherein an increment or decrement in the synaptic weight of interest due to spikes in the first neuron causing spikes in the second neuron is set to a constant value that is a multiplier on a change in the synaptic weight of interest.

14. The method as set forth in claim 8 , wherein tonic and phasic inputs are used to stabilize the synaptic weight of interest.

15. A computer program product for estimating conditional probabilities for operation of a mobile device, the computer program product comprising:
a non-transitory computer-readable medium having executable instructions encoded thereon, such that upon execution of the instructions by one or more processors, the one or more processors perform operations of:
inputting a first input data stream obtained from a first mobile device sensor and a second input data stream obtained from a second mobile device sensor into a neuronal network comprising a plurality of neurons, wherein the first and second input data streams are converted into variable spiking rates of a first neuron and a second neuron;
learning a conditional probability between the first input data stream and the second input data stream,
wherein a synaptic weight of interest between the first neuron and the second neuron converges to a fixed-point value corresponding to the conditional probability;
based on the conditional probability and a new input data stream, estimating a probability of an event; and
based on the probability of the event, causing the mobile device to perform a mobile device operation.

16. The computer program product as set forth in claim 15 , wherein the synaptic weight of interest is updated each time either of the first and second neurons spike according to spike-timing dependent plasticity (STDP), causing a corresponding update in the conditional probability such that the conditional probability is adapted in real-time.

17. The computer program product as set forth in claim 16 , wherein all synaptic connections between neurons in the neuronal network have a predetermined delay, and wherein all synaptic weights besides the synaptic weight of interest are fixed at a value such that only the synaptic weight of interest is updated according to STDP.

18. The computer program product as set forth in claim 17 , wherein an increment or decrement in the synaptic weight of interest due to spikes in the first neuron causing spikes in the second neuron is set to a constant value that is a multiplier on a change in the synaptic weight of interest.

19. The computer program product as set forth in claim 15 , wherein tonic and phasic inputs are used to stabilize the synaptic weight of interest.

20. A neuromorphic hardware chip for estimating conditional probabilities for a mobile device operation, the neuromorphic hardware chip performing operations of:
inputting a first input data stream obtained from a first mobile device sensor and a second input data stream obtained from a second mobile device sensor into a neuronal networ

Assignees

• Hrl Lab Llc

Inventors

• Jammalamadaka Aruna
• Stepp Nigel D

Classifications

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G06N3/08 Primary

  Learning methods · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G08G1/166

  for active traffic, e.g. moving vehicles, pedestrians, bikes · CPC title