Sequence-based anomaly detection with hierarchical spiking neural networks

Patent metadata

Field	Value
Publication number	US-12499973-B2
Application number	US-202217890843-A
Country	US
Kind code	B2
Filing date	Aug 18, 2022
Priority date	Jun 10, 2019
Publication date	Dec 16, 2025
Grant date	Dec 16, 2025

Abstract

Anomaly detection for streaming data is provided. A spiking neural network receives inputs of streaming data, wherein each input is contained within a number of neighborhoods and converts the inputs into phase-coded spikes. A median value of each input is calculated for each size neighborhood containing the input, and an absolute difference of each input from its median value is calculated for each size neighborhood. From the absolute differences, a median absolute difference (MAD) value of each input is calculated for each size neighborhood. It is determined whether the MAD value for any size neighborhood exceeds a respective threshold. If the MAD value exceeds its threshold, an anomaly indication is output for the input. If none of the MAD values for the neighborhoods exceeds its threshold, a normal indication is output for the input.

First claim

What is claimed is:

1 . A computer-implemented method of anomaly detection for streaming data, the method comprising:
using a number of processors to implement a spiking neural network that performs the steps of:
receiving inputs of streaming data, wherein each input is contained within a number of neighborhoods, wherein the neighborhoods comprise a number of increasingly larger symmetrical numbers of inputs preceding and following the input;
converting the inputs into phase-coded spikes;
calculating, from the phase-coded spikes, a median value of each input for each size neighborhood containing the input;
calculating an absolute difference of each input from its median value for each size neighborhood containing the input;
calculating, from the absolute differences, a median absolute difference (MAD) value of each input for each size neighborhood containing the input;
determining for each input whether the MAD value for any size neighborhood containing the input exceeds a respective threshold, wherein;
responsive to a determination that the MAD value of one or more neighborhoods exceeds its threshold, outputting an anomaly indication for the input; and
responsive to a determination that none of the MAD values for the neighborhoods exceeds its threshold, outputting a normal indication for the input.

2 . The method of claim 1 , further comprising normalizing the MAD values.

3 . The method of claim 1 , further comprising routing the determination of whether the MAD value for any size neighborhood exceeds a respective threshold back as input for computing the median value of each input for each neighborhood.

4 . The method of claim 1 , wherein phase-coding represents a neuron spiking delay for a specified length of time to allow processing the inputs in ascending order.

5 . The method of claim 1 , wherein the streaming data comprises one of:
sequential data; or time-series data.

6 . The method of claim 1 , wherein the inputs comprise a DNA sequence.

7 . The method of claim 6 , wherein detection of an anomaly indicates a DNA edit.

8 . The method of claim 1 , wherein the spiking neural network comprises leaky integrate-and-fire neurons.

9 . A system for anomaly detection for streaming data, the system comprising:
a storage device configured to store program instructions; and one or more processors operably connected to the storage device and configured to execute the program instructions to implement a spiking neural network that causes the system to:
receive inputs of streaming data, wherein each input is contained within a number of neighborhoods, wherein the neighborhoods comprise a number of increasingly larger symmetrical numbers of inputs preceding and following the input;
convert the inputs into phase-coded spikes;
calculate, from the phase-coded spikes, a median value of each input for each size neighborhood containing the input;
calculate an absolute difference of each input from its median value for each size neighborhood containing the input;
calculate, from the absolute differences, a median absolute difference (MAD) value of each input for each size neighborhood containing the input;
determine for each input whether the MAD value for any size neighborhood containing the input exceeds a respective threshold, wherein;
responsive to a determination that the MAD value of one or more neighborhoods exceeds its threshold, output an anomaly indication for the input; and
responsive to a determination that none of the MAD values for the neighborhoods exceeds its threshold, output a normal indication for the input.

10 . The system of claim 9 , wherein the spiking neural network normalizes the MAD values.

11 . The system of claim 9 , wherein the spiking neural network further routes the determination of whether the MAD value for any size neighborhood exceeds a respective threshold back as input for computing the median value of each input for each neighborhood.

12 . The system of claim 9 , wherein phase-coding represents a neuron spiking delay for a specified length of time to allow processing the inputs in ascending order.

13 . The system of claim 9 , wherein the streaming data comprises one of:
sequential data; or time-series data.

14 . The system of claim 9 , wherein the inputs comprise a DNA sequence.

15 . The system of claim 14 , wherein detection of an anomaly indicates a DNA edit.

16 . The system of claim 9 , wherein the spiking neural network comprises leaky integrate-and-fire neurons.

17 . A computer program product for anomaly detection for streaming data, the computer program product comprising:
a computer-readable storage medium having program instructions embodied thereon to implement a spiking neural network that performs the steps of:
receiving inputs of streaming data, wherein each input is contained within a number of neighborhoods, wherein the neighborhoods comprise a number of increasingly larger symmetrical numbers of inputs preceding and following the input;
converting the inputs into phase-coded spikes;
calculating, from the phase-coded spikes, a median value of each input for each size neighborhood containing the input;
calculating an absolute difference of each input from its median value for each size neighborhood containing the input;
calculating, from the absolute differences, a median absolute difference (MAD) value of each input for each size neighborhood containing the input;
determining for each input whether the MAD value for any size neighborhood containing the input exceeds a respective threshold, wherein;
responsive to a determination that the MAD value of one or more neighborhoods exceeds its threshold, outputting an anomaly indication for the input; and
responsive to a determination that none of the MAD values for the neighborhoods exceeds its threshold, outputting a normal indication for the input.

18 . The computer program product of claim 17 , further comprising instructions for normalizing the MAD values.

19 . The computer program product of claim 17 , further comprising instructions for routing the determination of whether the MAD value for any size neighborhood exceeds a respective threshold back as input for computing the median value of each input for each neighborhood.

20 . The computer program product of claim 17 , wherein phase-coding represents a neuron spiking delay for a specified length of time to allow processing the inputs in ascending order.

21 . The computer program product of claim 17 , wherein the streaming data comprises one of:
sequential data; or time-series data.

22 . The computer program product of claim 17 , wherein the inputs comprise a DNA sequence.

23 . The computer program product of claim 22 , wherein detection of an anomaly indicates a DNA edit.

24 . The computer program product of claim 17 , wherein the spiking neural network comprises leaky integrate-and-fire neurons.

Assignees

• Nat Tech & Eng Solutions Sandia Llc

Inventors

• Verzi Stephen Joseph
• Vineyard Craig Michael
• Aimone James Bradley

Classifications

• G06N3/049 Primary

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

• G06N3/08

  Learning methods · CPC title

• G16B40/20

  Supervised data analysis · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/045

  Combinations of networks · CPC title