Model reduction and training efficiency in computer-based reasoning and artificial intelligence systems
Techniques are provided herein for creating well-balanced computer-based reasoning systems and using those to control systems. The techniques include receiving a request to determine whether to use one or more particular data elements, features, cases, etc. in a computer-based reasoning model (e.g., as data elements, cases or features are being added, or as part of pruning existing features or cases). Conviction measures are determined and inclusivity conditions are tested. The result of comparing the conviction measure can be used to determine whether to include or exclude the feature, case, etc. in the model and/or whether there are anomalies in the model. A controllable system may then be controlled using the computer-based reasoning model. Examples controllable systems include self-driving cars, image labeling systems, manufacturing and assembly controls, federated systems, smart voice controls, automated control of experiments, energy transfer systems, health care systems, cybersecurity systems, and the like.
1. A method comprising:
receiving a request to determine whether to one or more particular data elements in a computer-based reasoning model meet inclusivity conditions;
determining one or more conviction scores for the one or more particular data elements, wherein:
determining for each of the one or more particular data elements, one or more conviction scores, wherein determining one or more conviction scores for the one or more particular data elements comprises determining a familiarity conviction score for the one or more particular data elements and determining a distance contribution score for the one or more particular data elements;
wherein:
the familiarity conviction score is a measure of how much the one or more particular data elements distort a model, and
the distance contribution score is a locally weighted expected value of the distance from one point to its nearest neighbors;
determining whether the one or more conviction scores meet one or more inclusivity conditions;
in response to determining that the one or more conviction scores meet the one or more inclusivity conditions:
including the one or more particular data elements in the computer-based reasoning model when the inclusivity conditions comprise an inclusion condition;
excluding the one or more particular data elements in the computer-based reasoning model when the inclusivity conditions comprise an exclusion condition;
causing, with a control system, control of a controllable system with the computer-based reasoning model,
wherein determining whether the one or more conviction scores meet the inclusivity conditions comprises determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is beyond a first threshold and the familiarity conviction score is beyond a second threshold,
wherein the method is performed on one or more computing devices.
2. The method of claim 1 , wherein determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is beyond the first threshold and the familiarity conviction score is beyond the second threshold comprises determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is below the first threshold and the familiarity conviction score is above the second threshold.
3. The method of claim 1 , wherein determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is beyond the first threshold and the familiarity conviction score is beyond the second threshold comprises determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is below the first threshold and the familiarity conviction score is below the second threshold.
4. The method of claim 1 , wherein determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is beyond the first threshold and the familiarity conviction score is beyond the second threshold comprises determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is above the first threshold and the familiarity conviction score is above the second threshold.
5. The method of claim 1 , wherein determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is beyond the first threshold and the familiarity conviction score is beyond the second threshold comprises determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is above the first threshold and the familiarity conviction score is below the second threshold.
6. The method of claim 1 ,
wherein receiving the request comprises receiving a request to reduce the computer-based reasoning model to a particular size;
and the method further comprises:
determining a number of data elements to exclude in the computer-based reasoning model to reduce the computer-based reasoning model to a particular size;
determining a subset of data elements to exclude in the computer-based reasoning model based at least in part on the one or more conviction scores for data elements in the computer-based reasoning model; and
excluding the subset of data elements from the computer-based reasoning model to reduce the size of the computer-based reasoning model to the particular size.
7. The method of claim 1 , further comprising:
initially receiving the one or more particular data elements as part of training for the computer-based reasoning model;
in response to determining that the one or more conviction scores meet the inclusion condition, sending an indication to a trainer associated with the training for the computer-based reasoning model to continue to train related to the one or more particular data elements;
in response to determining that the one or more conviction scores meet the exclusion condition, sending the indication to the trainer associated with the training for the computer-based reasoning model that training is no longer needed related to the one or more particular data elements.
8. The method of claim 1 , further comprising:
receiving a request for an action to take in a current context;
determining the action to take based on comparing the current context to contexts associated with cases in the computer-based reasoning model; and
responding to the request for the action to take with the determined action.
9. The method of claim 8 , further comprising:
receiving an indication that there was an anomaly associated with the determined action;
removing one or more cases associated with the determined action from the computer-based reasoning model.
10. The method of claim 1 , further comprising:
continuing to determine the one or more conviction scores for new cases and including or excluding those cases based on whether the one or more conviction scores meet the inclusivity conditions until a termination condition for inclusion or exclusion is met.
11. A system for executing instructions, wherein said instructions are instructions which, when executed by one or more computing devices, cause performance of a process including:
receiving a request to determine whether to one or more particular data elements in a computer-based reasoning model meet inclusivity conditions;
determining one or more conviction scores for the one or more particular data elements, wherein:
determining for each of the one or more particular data elements, one or more conviction scores, wherein determining one or more conviction scores for the one or more particular data elements comprises determining a familiarity conviction score for the one or more particular data elements and determining a distance contribution score for the one or more particular data elements;
wherein:
the familiarity conviction score is a measure of how much the one or more particular data elements distort a model, and
the distance contribution score is a locally weighted expected value of the distance from one point to its nearest neighbors;
determining whether the one or more conviction scores meet one or more inclusivity conditions;
in response to determining that the one or more conviction scores meet the one or more inclusivity conditions:
including the one or more particular data elements in the computer-based reasoning model when the inclusivity conditions comprise an inclusion condition;
excluding the one or more particular data elements in the computer-based reasoning model when the inclusivity conditions comprise an exclusion condition;
causing, with a control system, control of a controllable system with the computer-based reasoning model,
wherein determining whether the one or more conviction scores meet the inclusivity conditions comprises determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is beyond a first threshold and the familiarity conviction score is beyond a second threshold,
wherein the method is performed on one or more computing devices.
12. The system of claim 11 , wherein determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is beyond the first threshold and the familiarity conviction score is beyond the second threshold comprises determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is below the first threshold and the familiarity conviction score is above the second threshold.
13. The system of claim 11 , wherein determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is beyond the first threshold and the familiarity conviction score is beyond the second threshold comprises determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is below the first threshold and the familiarity conviction score is below the second threshold.
14. The system of claim 11 , wherein determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is beyond the first threshold and the familiarity conviction score is beyond the second threshold comprises determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is above the first threshold and the familiarity conviction score is above the second threshold.
15. The system of claim 11 , wherein determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is beyond the first threshold and the familiarity conviction score is beyond the second threshold comprises determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is above the first threshold and the familiarity conviction score is below the second threshold.
16. The system of claim 11 ,
wherein receiving the request comprises receiving a request to reduce the computer-based reasoning model to a particular size;
and the process further comprises:
determining a number of data elements to exclude in the computer-based reasoning model to reduce the computer-based reasoning model to a particular size;
determining a subset of data elements to exclude in the computer-based reasoning model based at least in part on the one or more conviction scores for data elements in the computer-based reasoning model; and
excluding the subset of data elements from the computer-based reasoning model to reduce the size of the computer-based reasoning model to the particular size.
17. The system of claim 11 , the process further comprising:
initially receiving the one or more particular data elements as part of training for the computer-based reasoning model;
in response to determining that the one or more conviction scores meet the inclusion condition, sending an indication to a trainer associated with the training for the computer-based reasoning model to continue to train related to the one or more particular data elements;
in response to determining that the one or more conviction scores meet the exclusion condition, sending the indication to the trainer associated with the training for the computer-based reasoning model that training is no longer needed related to the one or more particular data elements.
18. The system of claim 11 , the process further comprising:
receiving a request for an action to take in a current context;
determining the action to take based on comparing the current context to contexts associated with cases in the computer-based reasoning model; and
responding to the request for the action to take with the determined action.
19. The system of claim 18 , the process further comprising:
receiving an indication that there was an anomaly associated with the determined action;
removing one or more cases associated with the determined action from the computer-based reasoning model.
20. A non-transitory computer readable medium storing instructions which, when executed by one or more computing devices, cause the one or more computing devices to perform a process of:
receiving a request to determine whether to one or more particular data elements in a computer-based reasoning model meet inclusivity conditions;
determining one or more conviction scores for the one or more particular data elements, wherein:
determining for each of the one or more particular data elements, one or more conviction scores, wherein determining one or more conviction scores for the one or more particular data elements comprises determining a familiarity conviction score for the one or more particular data elements and determining a distance contribution score for the one or more particular data elements;
wherein:
the familiarity conviction score is a measure of how much the one or more particular data elements distort a model, and
the distance contribution score is a locally weighted expected value of the distance from one point to its nearest neighbors;
determining whether the one or more conviction scores meet one or more inclusivity conditions;
in response to determining that the one or more conviction scores meet the one or more inclusivity conditions:
including the one or more particular data elements in the computer-based reasoning model when the inclusivity conditions comprise an inclusion condition;
excluding the one or more particular data elements in the computer-based reasoning model when the inclusivity conditions comprise an exclusion condition;
causing, with a control system, control of a controllable system with the computer-based reasoning model,
wherein determining whether the one or more conviction scores meet the inclusivity conditions comprises determining that the one or more particular data elements meet the inclusion condition when the distance contribution score is beyond a first threshold and the familiarity conviction score is beyond a second threshold,
wherein the method is performed on one or more computing devices.