SPIKING DYNAMICAL NEURAL NETWORK FOR PARALLEL PREDICTION OF MULTIPLE TEMPORAL EVENTS

• US 20100179935A1
• Filed: 01/13/2009
• Published: 07/15/2010
• Est. Priority Date: 01/13/2009
• Status: Abandoned Application

• Current Assignee: GM Global Technology Operations LLC (General Motors Company)
• Original Assignee: GM Global Technology Operations Incorporated (Motors Liquidation Co. GUC Trust)
• Inventors: SRINIVASA, NARAYAN, Barajas, Leandro G., Cho, Youngkwan
• Application Number:
  US12/353,031
• Publication Number:
  US 20100179935A1
• Time in Patent Office: Days
• US Class Current: 706/21
• CPC Class Codes: G06N 3/049 Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs

Abstract

A system and method for determining events in a system or process, such as predicting fault events. The method includes providing data from the process, pre-processing data and converting the data to one or more temporal spike trains having spike amplitudes and a spike train length. The spike trains are provided to a dynamical neural network operating as a liquid state machine that includes a plurality of neurons that analyze the spike trains. The dynamical neural network is trained by known data to identify events in the spike train, where the dynamical neural network then analyzes new data to identify events. Signals from the dynamical neural network are then provided to a readout network that decodes the states and predicts the future events.

Claims

1. A method for determining temporal events, said method comprising:

providing data from a particular process;

converting the data to a temporal spike train having spike amplitudes and a spike train length;

training a dynamical neural network including a plurality of neurons to identify events;

providing the spike train to the trained dynamical neural network to analyze the spike train and predict events in the spike train; and

providing signals from the dynamical neural network to a readout device that identifies whether an event may occur.

2. The method according to claim 1 wherein converting the data to a spike train includes employing a class-based encoding scheme.

3. The method according to claim 1 wherein converting the data to a spike train includes employing a data-based encoding scheme.

4. The method according to claim 1 wherein converting the data to a spike train includes employing a space encoding scheme.

5. The method according to claim 1 wherein converting the data to a spike train includes employing a frequency-based encoding scheme.

6. The method according to claim 1 wherein the dynamical neural network operates as a liquid state machine.

7. The method according to claim 1 wherein the plurality of neurons include excitatory neurons and inhibitory neurons.

8. The method according to claim 7 wherein the ratio of excitatory neurons to inhibitory neurons is about 20% excitatory neurons and about 80% inhibitory neurons.

9. The method according to claim 1 wherein the method provides a parallel prediction of multiple temporal events simultaneously from a plurality of input spike trains.

10. The method according to claim 1 wherein the dynamical neural network is trained using a semi-supervised learning process.

11. The method according to claim 1 further comprising processing the data including sorting the data and classifying the data.

12. The method according to claim 1 wherein the method provides a prediction of temporal faults in a manufacturing process.

13. A method for providing a parallel prediction of multiple temporal fault events in a manufacturing process, said method comprising:

providing data from a particular process;

pre-processing the data to sort and classify the data;

converting the data to a plurality of temporal spike trains each having spike amplitudes and a spike train length;

training a dynamical neural network operating as a liquid state machine including a plurality of neurons to recognize fault events using a supervisory learning process;

providing the spike trains to the dynamical neural network to analyze the spike trains and predict fault events in the spike trains; and

providing signals from the dynamic neural network to a readout device that identifies whether a fault event may occur.

14. The method according to claim 13 wherein converting the data to a plurality of spike trains includes employing an encoding scheme from the group consisting of space encoding, frequency-based encoding, class-based encoding and data-based encoding.

15. The method according to claim 13 wherein the plurality of neurons include excitatory neurons and inhibitory neurons.

16. The method according to claim 15 wherein the ratio of excitatory neurons to inhibitory neurons is about 20% excitatory neurons and about 80% inhibitory neurons.

17. A method for providing a parallel prediction of multiple temporal fault events in a manufacturing process, said method comprising:

providing data from a particular process;

converting the data to a plurality of temporal spike trains each having spike amplitudes and a spike train length;

training a dynamical neural network operating as a liquid state machine including a plurality of neurons to recognize fault events; and

providing the spike trains to the dynamical neural network to analyze the spike trains and predict fault events in the spike trains.

18. The method according to claim 17 wherein converting the data to a plurality of spike trains includes employing an encoding scheme from the group consisting of space encoding, frequency-based encoding, class-based encoding and data-based encoding.

19. The method according to claim 17 wherein the plurality of neurons include excitatory neurons and inhibitory neurons.

20. The method according to claim 17 wherein the dynamical neural network is trained using a semi-supervised learning process.