Adaptive continuous machine learning by uncertainty tracking

1. A method of modulating a machine learning system configured to provide an output in response to an input data item, the machine learning system comprising a processor and a memory coupled to the processor, the memory storing the machine learning system configured to provide the output in response to the input data item and instructions, the method comprising:

determining, by the processor, a plurality of different kinds of uncertainty measures associated with the output, wherein two or more than two uncertainty measures of the plurality of different kinds of uncertainty measures are associated with a different hypothesis type of the machine learning system;

simultaneously comparing, by the processor, each uncertainty measure of the plurality of different kinds of uncertainty measures to a corresponding threshold of a plurality of thresholds;

determining, by the processor, that a sequential presentation of variations on the input data item cause the machine learning system to produce different outputs based on the simultaneous comparison;

adjusting, by the processor, a hypothesis type of the machine learning system, wherein the adjusted hypothesis type is associated with the input data item; and

adjusting, by the processor, a confidence threshold for an uncertainty measure of the plurality of different kinds of uncertainty measures.

2. The method of claim 1 , wherein a first uncertainty measure of the plurality of different kinds of uncertainty measures comprises whether there is a mismatch between the input data item and the output.

3. The method of claim 1 , wherein a second uncertainty measure of the plurality of different kinds of uncertainty measures comprises a similarity value between the input data item and a learned template for the output.

4. The method of claim 1 , wherein the a third uncertainty measure of the plurality of different kinds of uncertainty measures comprises a degree of relation between the input data item and each potential output of the machine learning system.

5. The method of claim 1 , wherein the sequential presentation of variations on the input data item cause the machine learning system to produce different outputs.

6. The method of claim 1 , wherein a fourth uncertainty measure of the plurality of different kinds of uncertainty measures comprises a degree of relation between the input data item and each potential output of the machine learning system relative to predetermined expectations.

7. The method of claim 1 , wherein a fifth uncertainty measure of the plurality of different kinds of uncertainty measures comprises whether data at each data processing step of the machine learning system corresponds to a learned template for the output at that particular data processing step.

8. The method of claim 1 , wherein a sixth uncertainty measure of the plurality of different kinds of uncertainty measures comprises whether the output is within expectations given a current context or assigned environment in which the machine learning system is operating.

9. The method of claim 1 , wherein a seventh uncertainty measure of the plurality of different kinds of uncertainty measures comprises a degree of entropy in the input data item.

10. The method of claim 1 , wherein an eighth uncertainty measure of the plurality of different kinds of uncertainty measures comprises a degree of entropy at each data processing stage of the machine learning system.

11. The method of claim 1 , wherein a ninth uncertainty measure of the plurality of different kinds of uncertainty measures comprises a degree to which an operational condition has varied from a predetermined condition.

12. The method of claim 1 , wherein:

the machine learning system comprises a distributed machine learning system; and

a tenth uncertainty measure of the plurality of different kinds of uncertainty measures comprises a degree of agreement between each component of the distributed machine learning system.

13. The method of claim 1 , wherein an eleventh uncertainty measure of the plurality of different kinds of uncertainty measures comprises whether the output is mutually exclusive with other elements of an environment within which the machine learning system is operating.

14. The method of claim 1 , wherein the machine learning system comprises an artificial neural network.

15. The method of claim 1 , wherein the machine learning system comprises a classifier.

16. The method of claim 1 , wherein the machine learning system comprises a recurrent neural network.

17. The method of claim 1 , wherein the machine learning system comprises a support vector machine.

18. The method of claim 1 , wherein the machine learning system comprises a machine learning algorithm or model.

19. The method of claim 1 , wherein adjusting the hypothesis type comprises adjusting internal representations of the machine learning system.

20. The method of claim 1 , further comprising:

determining, by the processor, that a third confidence level is below a respective confidence threshold; and

defining, by the processor, a new output type corresponding to an object defined by the input data item.

21. The method of claim 20 , further comprising:

providing, by the processor, a user prompt to apply a label to the new output type.

22. A computer system for modulating a machine learning system configured to provide an output in response to an input data item, the computer system comprising:

a processor; and

a memory coupled to the processor, the memory storing the machine learning system and instructions that, when executed by the processor, cause the computer system to:

determine a plurality of different kinds of uncertainty measures associated with the output, wherein two or more than two uncertainty measures of the plurality of different kinds of uncertainty measures are associated with a different hypothesis type of the machine learning system;

simultaneously compare each uncertainty measure of the plurality of different kinds of uncertainty measures to a corresponding threshold of a plurality of thresholds;

determine that a sequential presentation of variations on the input data item cause the machine learning system to produce different outputs based on the simultaneous comparison;

adjust a hypothesis type of the machine learning system, wherein the adjusted hypothesis type is associated with the input data item; and

adjust a confidence threshold for an uncertainty measure of the plurality of different kinds of uncertainty measures.

23. The computer system of claim 22 , wherein a first uncertainty measure of the plurality of different kinds of uncertainty measures comprises whether there is a mismatch between the input data item and the output.

24. The computer system of claim 22 , wherein a second uncertainty measure of the plurality of different kinds of uncertainty measures comprises a similarity value between the input data item and a learned template for the output.

25. The computer system of claim 22 , wherein a third uncertainty measure of the plurality of different kinds of uncertainty measures comprises a degree of relation between the input data item and each potential output of the machine learning system.

26. The computer system of claim 22 , wherein the sequential presentation of variations on the input data item cause the machine learning system to produce different outputs.

27. The computer system of claim 22 , wherein a fourth uncertainty measure of the plurality of different kinds of uncertainty measures comprises a degree of relation between the input data item and each potential output of the machine learning system relative to predetermined expectations.

28. The computer system of claim 22 , wherein a fifth uncertainty measure of the plurality of different kinds of uncertainty measures comprises whether data at each data processing step of the machine learning system corresponds to a learned template for the output at that particular data processing step.

29. The computer system of claim 22 , wherein a sixth uncertainty measure of the plurality of different kinds of uncertainty measures comprises whether the output is within expectations given a current context or assigned environment in which the machine learning system is operating.

30. The computer system of claim 22 , wherein a seventh uncertainty measure of the plurality of different kinds of uncertainty measures comprises a degree of entropy in the input data item.

31. The computer system of claim 22 , wherein an eighth uncertainty measure of the plurality of different kinds of uncertainty measures comprises a degree of entropy at each data processing stage of the machine learning system.

32. The computer system of claim 22 , wherein a ninth uncertainty measure of the plurality of different kinds of uncertainty measures comprises a degree to which an operational condition has varied from a predetermined condition.

33. The computer system of claim 22 , wherein:

the machine learning system comprises a distributed machine learning system; and

a tenth uncertainty measure of the plurality of different kinds of uncertainty measures comprises a degree of agreement between each component of the distributed machine learning system.

34. The computer system of claim 22 , wherein an eleventh uncertainty measure of the plurality of different kinds of uncertainty measures comprises whether the output is mutually exclusive with other elements of an environment within which the machine learning system is operating.

35. The computer system of claim 22 , wherein the machine learning system comprises an artificial neural network.

36. The computer system of claim 22 , wherein the machine learning system comprises a classifier.

37. The computer system of claim 22 , wherein the instructions stored by the memory, when executed by the processor, cause the computer system to adjust the hypothesis type by causing the computer system to adjust internal representations of the machine learning system.

38. The computer system of claim 22 , wherein the instructions stored by the memory, when executed by the processor, further cause the computer system to:

determine that a third confidence level is below a respective confidence threshold; and

define a new output type corresponding to an object defined by the input data item.

39. The computer system of claim 38 , wherein the instructions stored by the memory, when executed by the processor, further cause the computer system to:

provide a user prompt to apply a label to the new output type.