Method and system for graph level anomaly detection
A computer-implemented method of anomaly detection associated with graphical data includes receiving as input one or more input data sets, wherein the input data sets includes one or more graphs, utilizing a plurality of graph neural networks (GNNs) to identify an aggregate loss including a first loss and second loss associated with the input data set, wherein the aggregate loss is associated with embedding's of the GNNs, and outputting a notification associated with an anomaly detection when the first and second loss exceeds an anomaly-detection threshold.
1. A computer-implemented method of anomaly detection associated with graphical data, comprising:
receiving as input one or more input data sets, wherein the input data sets includes one or more graphs;
utilizing a plurality of graph neural networks (GNNs) to identify an aggregate loss including a first loss and second loss associated with the input data set, wherein the aggregate loss is associated with embeddings of the input graphs produced by the GNNs, wherein the first loss is associated with a Euclidean distance and a second loss is associated angles between embeddings of the input graphs;
training a plurality of feature extractors associated with the plurality of graph neural networks in response to the aggregate loss, wherein the plurality of feature extractors associated with the plurality of graph neural networks are trained jointly; and
outputting a notification associated with an anomaly detection when the first and second loss exceeds an anomaly-detection threshold.
2. The computer-implemented method of claim 1 , wherein the first loss is a one-class classification term.
3. The computer-implemented method of claim 1 , wherein the method further includes the step of intervening in response to the anomaly detection.
4. The computer-implemented method of claim 1 , wherein the method further includes the step of updating parameters associated with the GNNs and identifying one or more losses associated with the input data set in response to the falling below the convergence threshold.
5. The computer-implemented method of claim 1 , wherein the graph neural networks are trained graph neural networks.
6. The computer-implemented method of claim 1 , wherein the graphical data is associated with one or more images.
7. A computer-implemented method of anomaly detection associated with graphical data, comprising:
receiving as input one or more input data sets, wherein the input data sets include one or more graphs;
utilizing a plurality of graph neural networks to identify a loss associated with the input data set, wherein the loss utilizes one or more embeddings associated with the one or more graphs;
training a plurality of feature extractors associated with the plurality of graph neural networks based on the loss, wherein the plurality of feature extractors associated with the plurality of graph neural networks are trained jointly; and
outputting a trained graph neural network in response to the loss exceeding a convergence threshold.
8. The computer-implemented method of claim 7 , wherein the loss includes a first loss that is a one-class classification loss and a second loss that is a contrastive loss of the one or more embeddings.
9. The computer-implemented method of claim 7 , wherein the training includes jointly training the feature extractors.
10. The computer-implemented method of claim 7 , wherein the method further includes the step of updating parameters associated with the graph neural networks and identifying one or more losses associated with the input data set in response to the falling below the convergence threshold.
11. The computer-implemented method of claim 7 , wherein the loss includes a first loss that is a one-class classification loss.
12. The computer-implemented method of claim 7 , wherein the graphical data is associated with one or more pixels.
13. A system including a machine-learning network, comprising:
an input interface configured to receive input data including graphical representations;
a processor, in communication with the input interface, wherein the processor is programmed to:
receive as input one or more input data sets, wherein the input data sets include one or more graphs;
utilize a plurality of graph neural networks to identify a first loss and second loss associated with the input data set;
train a plurality of feature extractors associated with the plurality of graph neural networks in response to the first and second loss, wherein the plurality of feature extractors associated with the plurality of graph neural networks are trained jointly; and
output a trained plurality of graph neural networks in response to the first and second loss exceed a convergence threshold.
14. The system of claim 13 , wherein the second loss is a contrastive loss of embeddings.
15. The system of claim 13 , wherein the loss includes a first loss that is a one-class classification loss.
16. The system of claim 13 , wherein the processor is further programmed to update parameters associated with the graph neural networks when the first and second loss fall below the convergence threshold.
17. The system of claim 13 , wherein the graphical representations is associated with a plurality of nodes and edges and node features.
18. The system of claim 13 , wherein the graphical representations are associated with social network data or financial network data.
19. The system of claim 13 , wherein the graphical data is associated with one or more pixels.