Method and apparatus for performing anomaly detection using neural network

Patent metadata

Field	Value
Publication number	US-11727279-B2
Application number	US-202016781328-A
Country	US
Kind code	B2
Filing date	Feb 4, 2020
Priority date	Jun 11, 2019
Publication date	Aug 15, 2023
Grant date	Aug 15, 2023

Abstract

A method and apparatus for performing anomaly detection by using a neural network are provided. The apparatus is configured to extract input features of an input data signal, obtain output features of the neural network by processing the input features through the neural network, obtain an error based on the input features and the output features, and determine whether the input data signal indicates an abnormal signal based on the error and a threshold.

First claim

What is claimed is:

1. A method of operating an anomaly detector including processing circuitry comprising a neural network, the neural network including layers of an encoder and a decoder, the method comprising:
extracting, by the processing circuitry, input features of an input data signal;
processing, by the processing circuitry, the input features using the neural network such that output features of the neural network corresponding to an output of the decoder are obtained;
performing, by the processing circuitry, unsupervised learning on the neural network based on the output features such that the neural network is trained to differentiate an abnormal signal from a normal signal;
obtaining, by the processing circuitry, an error based on the input features and the output features;
determining, by the processing circuitry, whether the input data signal indicates the abnormal signal or the normal signal based on a comparison of the error and a threshold; and
outputting, by the processing circuitry, information indicating that the abnormal signal is detected based on a determination that the input data signal indicates the abnormal signal,
wherein the neural network further includes an input layer and an output layer, and
wherein the performing of the unsupervised learning includes updating weights of the input layer and the output layer through online learning when the input data signal indicates the normal signal such that a difference between the input features and the output features decreases.

2. The method of claim 1 , wherein the neural network includes,
an autoencoder that configures the layers of the encoder and the decoder based on the input layer, at least one hidden layer, and the output layer.

3. The method of claim 2 , wherein a dimension of the input layer and a dimension of the output layer are greater than a dimension of the hidden layer.

4. The method of claim 2 , wherein the obtaining the error includes,
calculating a mean squared error (MSE) between the input features and the output features.

5. The method of claim 1 , wherein the performing of the unsupervised learning includes,
updating the weights by performing learning so that the error decreases according to gradient descent.

6. The method of claim 5 , wherein the performing of the unsupervised learning includes,
updating the weights through a predetermined number of batch learnings.

7. A non-transitory computer-readable recording medium having recorded thereon a program that, when executed by processing circuitry of an apparatus, causes the apparatus to perform the method of claim 1 .

8. An apparatus for performing anomaly detection by using a neural network, the apparatus comprising:
a memory storing at least one program; and
processing circuitry configured to perform the anomaly detection by executing the at least one program,
wherein the processing circuitry is configured to:
extract input features of an input data signal,
process the input features by a neural network including an input layer, an output layer, and layers of an encoder and a decoder to obtain output features of the neural network corresponding to an output of the decoder,
obtain an error based on the input features and the output features,
determine whether the input data signal indicates an abnormal signal based on a comparison of the error and a threshold,
output information indicating that the abnormal signal is detected based on determining that the input data signal indicates the abnormal signal, and
perform unsupervised learning on the neural network based on the output features,

wherein the performing of the unsupervised learning includes updating weights of the input layer and the output layer through online learning when the input data signal indicates a normal signal such that a difference between the input features and the output features decreases.

9. The apparatus of claim 8 , wherein the neural network includes,
an autoencoder that configures the layers of the encoder and the decoder based on the input layer, at least one hidden layer, and the output layer.

10. The apparatus of claim 9 , wherein a dimension of the input layer and a dimension of the output layer are greater than a dimension of at least one of the at least one hidden layer.

11. The apparatus of claim 9 , wherein the obtaining the error includes,
calculating a mean squared error (MSE) between the input features and the output features.

12. The apparatus of claim 8 , wherein the updating, through the unsupervised learning, includes,
updating the weights by performing learning so that the error decreases according to gradient descent.

13. The apparatus of claim 12 , wherein the processing circuitry is further configured to,
update the weights through a predetermined number of batch learnings.

14. The apparatus of claim 8 , wherein the processing circuitry includes,
an encoder layer calculation circuit that is configured to perform calculation for a layer of the encoder, and
a decoder layer calculation circuit that is configured to perform calculation for a layer of the decoder.

15. The apparatus of claim 8 , wherein the processing circuitry includes,
a layer calculation circuit configured to perform, with one circuit configuration, calculation for a layer of the encoder and calculation for a layer of the decoder.

Assignees

• Samsung Electronics Co Ltd
• Postech Res & Business Dev Found

Inventors

• Kim Hyunsoo
• Sim Jaeyoon
• Park Jaehan
• Son Hyunwoo
• Kim Sangjoon

Classifications

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/0455 Primary

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06F2218/12

  Classification; Matching · CPC title

• G06F18/2413

  based on distances to training or reference patterns · CPC title

• G06N3/084

  Backpropagation, e.g. using gradient descent · CPC title