Machine learning model explanation apparatus and methods

1. An explanation apparatus comprising:

processing circuitry configured to:

access a source instance which has been classified as a result of a classification decision of the source instance by a machine learning model;

create associations of the source instance with a plurality of training instances; and

process the associations of the source instance and the training instances to identify a first subset of the training instances which have less relevance to the classification decision of the source instance by the machine learning model compared with a second subset of the training instances;

an interface configured to communicate information to a user; and

wherein the processing circuitry is configured to control the user interface to communicate the second subset of the training instances to the user as evidence to explain the classification decision of the source instance by the machine learning model.

2. The apparatus of claim 1 wherein some of the training instances are parent instances and some of the training instances are descendant instances of the parent instances, and wherein the processing circuitry is configured to include selected ones of the descendant instances in the first subset of the training instances as a result of the selected descendant instances having parent instances which have a different classification than the classification of the source instance.

3. The apparatus of claim 2 wherein the processing circuitry is to include additional ones of the descendant instances in the first subset of the training instances as a result of the additional descendant instances being non-leaf nodes which each have no descendant instances that have a classification which is different than the classification of the source instance.

4. The apparatus of claim 1 wherein the source instance accessed by the processing circuitry comprises a model representation of raw data of the source instance, and wherein the processing circuitry is configured to access model representations of raw data of the training instances and to process the model representation of the source instance and the model representations of the training instances to create the associations of the source instance with the training instances.

5. The apparatus of claim 4 wherein the processing circuitry is configured to process the raw data of the source instance and the training instances to generate the model representations of the source instance and the training instances.

6. The apparatus of claim 1 wherein the processing circuitry is configured to create a topological structure including the source instance, the training instances and the associations of the source instance with the training instances.

7. The apparatus of claim 6 wherein the topological structure comprises a plurality of edges between the source instance and the training instances.

8. The apparatus of claim 6 wherein the topological structure is a k-nearest neighbor graph.

9. The apparatus of claim 6 wherein the processing circuitry is configured to use the topological structure to create a shortest-path tree from the source instance to the training instances and which includes the associations of the source instance with the training instances, and to process the associations of the source instance with the training instances of the shortest-path tree to identify the training instances of the first subset.

10. The apparatus of claim 1 wherein the processing circuitry is configured to control the interface to display the second subset of the training instances associated with the source instance to communicate the second subset of the training instances to the user.

11. The apparatus of claim 1 wherein the processing circuitry is configured to access classifications of the training instances.

12. The apparatus of claim 11 wherein the processing circuitry uses classifications of the source instance and the training instances to create the associations of the source instance with the training instances.

13. The apparatus of claim 11 wherein the processing circuitry is configured to use the classifications of the training instances to identify the first subset of the training instances.

14. The apparatus of claim 11 wherein the second subset comprises the training instances along a plurality of shortest paths between the source instance and the training instances having different classifications from the classification of the source instance.

15. The apparatus of claim 11 wherein the processing circuitry is configured to execute the machine learning model to classify the source instance.

16. The apparatus of claim 1 wherein the training instances were used to train the machine learning model prior to the classification of the source instance.

17. The apparatus of claim 1 wherein each of the first and second subsets of the training instances comprise plural ones of the training instances.

18. The apparatus of claim 1 wherein the processing circuitry is configured to control the user interface to communicate raw data that is associated with the training instances of the second subset to the user as additional evidence to explain the classification decision of the source instance by the machine learning model.

19. The apparatus of claim 1 wherein the source instance is unclassified prior to the classification by the machine learning model.

20. An explanation apparatus comprising:

processing circuitry configured to:

generate a k-nearest neighbors graph comprising a source instance and a plurality of training instances, and wherein the source instance and the training instances have a plurality of different classifications;

create a shortest-path tree from the source instance to the training instances;

first remove each of the training instances from the shortest-path tree which has a parent training instance in the shortest-path tree with a classification different than the classification of the source instance;

second remove each of the training instances from the shortest-path tree which is a non-leaf node in the shortest-path tree and which has no descendant instances which have a classification different than the classification of the source instance; and

after the first and second removings, initiate communication of at least one of the remaining training instances of the shortest-path tree to a user as evidence to explain a classification decision of the source instance by a machine learning model.

21. The apparatus of claim 20 wherein the processing circuitry is configured to use the machine learning model to generate model representations of the source instance and the training instances, and to use the model representations of the source instance and the training instances to generate the k-nearest neighbors graph.

22. The apparatus of claim 20 further comprising a user interface, and wherein the processing circuitry is configured to initiate display of the at least one of the remaining training instances using the user interface to initiate the communication of the at least one remaining training instance to the user.

23. The apparatus of claim 20 further comprising a user interface, and wherein the processing circuitry is configured to initiate display of the remaining training instances associated with one another within the shortest-path tree to initiate the communication of the at least one remaining training instance to the user.