Deep learning device and system including the same
A deep learning device and system including the same is provided. The deep learning device comprising processing circuitry configured to determine whether a received image is abnormal using an anomaly detection model; merge at least some vectors extracted from the anomaly detection model; input, to a probability approximation model, principal components generated by a principal component analysis (PCA) to detect whether out of distribution (OOD) occurs in data of the received image; store a result of the determinations; and extract at least some the data in which the OOD occurs, as target labeling, using a target labeling extraction model when a rate of the data in which the OOD occurs is greater than or equal to a threshold value, wherein the anomaly detection model determines whether the received image is abnormal using the target labeling.
1. A deep learning device comprising:
processing circuitry configured to:
determine whether a received image is abnormal using an anomaly detection model,
merge at least some vectors extracted from the anomaly detection model,
perform a principal component analysis (PCA) on a result of the merging the at least some vectors,
input, to a probability approximation model, principal components generated by the PCA to detect whether out of distribution (OOD) occurs in data of the received image,
store, in a database, a result of the determination of whether the received image is abnormal and a result of determining whether the OOD occurs, and
extract at least some of the data in which the OOD occurs, as target labeling, using a target labeling extraction model when a rate of the data in which the OOD occurs is greater than or equal to a threshold value,
wherein the anomaly detection model determines whether the received image is abnormal using the target labeling.
2. The deep learning device of claim 1 , wherein the anomaly detection model includes at least one of a convolutional neural network (CNN), an artificial neural network (ANN), a dense convolutional network (DenseNet), or a scaled CNN.
3. The deep learning device of claim 1 , wherein the probability approximation model includes at least one of a Gaussian Mixture Model (GMM), a histogram-based model, a kernel density estimation model, a Parzen window estimation model, a k-nearest neighbor estimation model, or a k-means clustering model.
4. The deep learning device of claim 1 , wherein the processing circuitry is further configured to check, in real time, whether the rate of the data in which the OOD occurs is greater than or equal to the threshold value.
5. The deep learning device of claim 1 , wherein the processing circuitry is further configured to:
detect an optimal value for clustering between the principal components and the probability approximation model; and
detect whether the OOD occurs using at least some of the principal components which satisfy the optimal value and at least part of the probability approximation model which satisfy the optimal value.
6. The deep learning device of claim 5 , wherein the optimal value is calculated according to at least one of a Bayesian information criterion (BIC), a Dunn index, a silhouette score, or an Akaike information criterion (AIC).
7. The deep learning device of claim 1 , wherein the threshold value comprises a rate of the data in which the OOD occurs with respect to the image stored in the database when performance precision is a set value.
8. A deep learning system comprising:
an input unit configured to receive an image;
a processor configured to control the input unit; and
a deep learning device connected to the processor and configured to receive the image from the input unit, the deep learning device including processing circuitry and a database, the processing circuitry configured to
determine whether an image received from the input unit is abnormal using an anomaly detection model,
merge at least some vectors extracted from the anomaly detection model,
perform a principal component analysis (PCA) on a result of the merging the at least some vectors,
input, to a probability approximation model, principal components generated by the PCA to detect whether out of distribution (OOD) occurs in data of the image,
store a result of the determination of whether the image is abnormal and a result of determining whether the OOD occurs, and
extract at least some of the data in which the OOD occurs, as target labeling, using a target labeling extraction model when a rate of the data in which the OOD occurs is greater than or equal to a threshold value, and
the database is configured to store the result of the determination of whether the image is abnormal and the result of determining whether the OOD occurs,
wherein the anomaly detection model determines whether the image is abnormal using the target labeling.
9. The deep learning system of claim 8 , wherein the anomaly detection model includes at least one of a convolutional neural network (CNN), an artificial neural network (ANN), a dense convolutional network (DenseNet), or scaled CNN.
10. The deep learning system of claim 8 , wherein the probability approximation model includes at least one of a Gaussian Mixture Model (GMM), a histogram-based model, a kernel density estimation model, a Parzen window estimation model, a k-nearest neighbor estimation model or a k-means clustering model.
11. The deep learning system of claim 8 , wherein the processing circuitry is further configured to check, in real time, whether the rate of the data in which the OOD occurs is greater than or equal to the threshold value.
12. The deep learning system of claim 8 , wherein the processing circuitry is configured to:
detect an optimal value for clustering between the principal components and the probability approximation model; and
detect whether the OOD occurs using at least some of the principal components which satisfy an optical value and at least part of the probability approximation model which satisfy the optimal value.
13. The deep learning system of claim 12 , wherein the optimal value is calculated according to at least one of a Bayesian information criterion (BIC), a Dunn index, a silhouette score, or an Akaike information criterion (AIC).
14. The deep learning system of claim 8 , wherein the threshold value comprises a rate of the data in which the OOD occurs with respect to the image stored in the database when performance precision is a set value.
15. A deep learning system comprising:
an input unit configured to receive an image;
a processor configured to control the input unit; and
a deep learning device connected to the processor and configured to receive the image from the input unit, the deep learning device including processing circuitry configured to:
determine whether an image received from the input unit is abnormal using an anomaly detection model,
merge at least some vectors extracted from the anomaly detection model,
perform a principal component analysis (PCA) on a result of the merging the at least some vectors,
input, to a probability approximation model, principal components generated by the PCA to detect whether out of distribution (OOD) occurs in data of the image,
store, in a database, a result of the determining whether the image is abnormal and the result of detecting whether the OOD occurs, and
extract at least some pieces of data in which the OOD occurs using a target labeling extraction model when a rate of data in which the OOD occurs with respect to the image stored in the database is greater than or equal to a threshold value,
wherein the anomaly detection model determines whether the image is abnormal using the target labeling extraction model.
16. The deep learning system of claim 15 , wherein the anomaly detection model includes at least one of a convolutional neural network (CNN), an artificial neural network (ANN), a dense convolutional network (DenseNet), or scaled CNN.
17. The deep learning system of claim 15 , wherein the probability approximation model includes at least one of a Gaussian Mixture Model (GMM), a histogram-based model, a kernel density estimation model, a Parzen window estimation model, a k-nearest neighbor estimation model or a k-means clustering model.
18. The deep learning system of claim 15 , wherein the processing circuitry is further configured to check, in real time, whether the rate of the data in which the OOD occurs is greater than or equal to the threshold value.
19. The deep learning system of claim 15 , wherein the processing circuitry is configured to:
detect an optimal value for clustering between the principal components and the probability approximation model, and
detect whether the OOD occurs using at least some of the principal components which satisfy an optical value and at least part of the probability approximation model which satisfy the optimal value.
20. The deep learning system of claim 15 , wherein the threshold value comprises a rate of the data in which the OOD occurs with respect to the image stored in the database when performance precision is a set value.