Cross-sensor predictive inference

1. A computer-implemented method for cross-sensor predictive inference, the computer-implemented method comprising:

identifying a plurality of sensor input data objects comprising one or more image data objects;

determining a plurality of sensor feature data objects based at least in part on the plurality of sensor input data objects, wherein determining the plurality of sensor feature data objects comprises:

for each sensor input data object of the plurality of sensor input data objects, determining one or more desired observation metrics based at least in part on a feature definition model for the sensor input data object; and

for each desired observation metric of the one or more desired observation metrics that is associated with a sensor input data object of the plurality of sensor input data objects, determining a sensor feature data object of the plurality of sensor feature data object by processing the sensor input data object using a feature extraction model of a plurality of feature extraction models that is associated with the sensor input data object, wherein the plurality of feature extraction models comprises one or more image feature extraction models;

generating one or more cross-sensor predictions for a target predictive entity associated with the plurality of sensor input data objects by processing the plurality of sensor feature data objects using a cross-sensor predictive inference model; and

performing one or more prediction-based actions based at least in part on the one or more cross-sensor predictions.

2. The computer-implemented method of claim 1 , wherein the plurality of feature extraction models comprises a convolutional neural network.

3. The computer-implemented method of claim 1 , wherein the plurality of feature extraction models comprises a capsule-based machine learning model.

4. The computer-implemented method of claim 1 , wherein determining the one or more desired observation metrics for a sensor input data object of the plurality of sensor input data objects comprises:

identifying one or more potential observation metrics for the sensor input data object; and

determining the one or more desired observation metrics by selecting a desired subset of the one or more potential observation metrics based at least in part on the feature definition model for the sensor.

5. The computer-implemented method of claim 1 , wherein the feature definition model is defined based at least in part on one or more inferential properties of the target predictive entity.

6. The computer-implemented method of claim 1 , wherein the feature definition model is defined based at least in part on a feature confidence measure for each potential observation metric of the one or more potential observation metrics.

7. The computer-implemented method of claim 1 , wherein generating a first feature extraction model of the plurality of feature models that is associated with a first sensor data object of the plurality of sensor data objects comprises:

identifying a generic model and a local model, wherein the local model is generated based at least in part on local training data associated with a physical environment of the first sensor data object;

determining whether a local model confidence measure for the local model exceeds a model confidence threshold value;

responsive to determining that the local model confidence measure for the local model exceeds the model confidence threshold value, adopting the local model as the first feature extraction model; and

responsive to determining that the local model confidence measure for the local model fails to exceed the model confidence threshold value, adopting the generic model as the first feature extraction model.

8. The computer-implemented method of claim 7 , wherein the model confidence threshold value is determined based at least in part on a generic model confidence measure for the generic model.

9. The computer-implemented method of claim 1 , wherein generating a first feature extraction model of the plurality of feature models comprises:

identifying one or more model generating devices associated with the first feature extraction model;

receiving, from each model generating device of the one or more model generating devices, weight update data for the first feature extraction model; and

determining the first feature extraction model based at least in part on each weight update data received from a model generating device of the one or more model generating devices.

10. The computer-implemented method of claim 9 , wherein determining the first feature extraction model further comprises determining first feature extraction model based at least in part on each training intensity measure for a model generating device of the one or more model generating devices.

11. The computer-implemented method of claim 1 , wherein:

each sensor input data object of the plurality of sensor input data objects is associated with a sensor of a plurality sensors, and

each feature definition model for a sensor input data object of the plurality of sensor input data objects is determined based at least in part on the sensor of the plurality sensors that is associated with the sensor input data object.

12. An apparatus for cross-sensor predictive inference, the apparatus comprising at least one processor and at least one memory including program code, the at least one memory and the program code configured to, with the processor, cause the apparatus to at least:

identify a plurality of sensor input data objects comprising one or more image data objects;

determine a plurality of sensor feature data objects based at least in part on the plurality of sensor input data objects, wherein determining the plurality of sensor feature data objects comprises:

for each sensor input data object of the plurality of sensor input data objects, determining one or more desired observation metrics based at least in part on a feature definition model for the sensor input data object; and

for each desired observation metric of the one or more desired observation metrics that is associated with a sensor input data object of the plurality of sensor input data objects, determining a sensor feature data object of the plurality of sensor feature data object by processing the sensor input data object using a feature extraction model of a plurality of feature extraction models that is associated with the sensor input data object, wherein the plurality of feature extraction models comprise one or more image feature extraction models;

generate one or more cross-sensor predictions for a target predictive entity associated with the plurality of sensor input data objects by processing the plurality of sensor feature data objects using a cross-sensor predictive inference model; and

perform one or more prediction-based actions based at least in part on the one or more cross-sensor predictions.

13. The apparatus of claim 12 , wherein determining the one or more desired observation metrics for a sensor input data object of the plurality of sensor input data objects comprises:

identifying one or more potential observation metrics for the sensor input data object; and

determining the one or more desired observation metrics by selecting a desired subset of the one or more potential observation metrics based at least in part on the feature definition model for the sensor.

14. The apparatus of claim 12 , wherein generating a first feature extraction model of the plurality of feature models that is associated with a first sensor data object of the plurality of sensor data objects comprises:

identifying a generic model and a local model, wherein the local model is generated based at least in part on local training data associated with a physical environment of the first sensor data object;

determining whether a local model confidence measure for the local model exceeds a model confidence threshold value;

responsive to determining that the local model confidence measure for the local model exceeds the model confidence threshold value, adopting the local model as the first feature extraction model; and

responsive to determining that the local model confidence measure for the local model fails to exceed the model confidence threshold value, adopting the generic model as the first feature extraction model.

15. The apparatus of claim 12 , wherein generating a first feature extraction model of the plurality of feature models comprises:

identifying one or more model generating devices associated with the first feature extraction model;

receiving, from each model generating device of the one or more model generating devices, weight update data for the first feature extraction model; and

determining the first feature extraction model based at least in part on each weight update data received from a model generating device of the one or more model generating devices.

16. The apparatus of claim 12 , wherein:

each sensor input data object of the plurality of sensor input data objects is associated with a sensor of a plurality sensors, and

each feature definition model for a sensor input data object of the plurality of sensor input data objects is determined based at least in part on the sensor of the plurality sensors that is associated with the sensor input data object.

17. A computer program product for cross-sensor predictive inference, the computer program product comprising at least one non-transitory computer-readable storage medium having computer-readable program code portions stored therein, the computer-readable program code portions configured to:

identify a plurality of sensor input data objects comprising one or more image data objects;

determine a plurality of sensor feature data objects based at least in part on the plurality of sensor input data objects, wherein determining the plurality of sensor feature data objects comprises:

for each sensor input data object of the plurality of sensor input data objects, determining one or more desired observation metrics based at least in part on a feature definition model for the sensor input data object; and

for each desired observation metric of the one or more desired observation metrics that is associated with a sensor input data object of the plurality of sensor input data objects, determining a sensor feature data object of the plurality of sensor feature data object by processing the sensor input data object using a feature extraction model of a plurality of feature extraction models that is associated with the sensor input data object, wherein the plurality of feature extraction models comprises one or more image feature extraction models;

generate one or more cross-sensor predictions for a target predictive entity associated with the plurality of sensor input data objects by processing the plurality of sensor feature data objects using a cross-sensor predictive inference model; and

perform one or more prediction-based actions based at least in part on the one or more cross-sensor predictions.

18. The computer program product of claim 17 , wherein determining the one or more desired observation metrics for a sensor input data object of the plurality of sensor input data objects comprises:

identifying one or more potential observation metrics for the sensor input data object; and

determining the one or more desired observation metrics by selecting a desired subset of the one or more potential observation metrics based at least in part on the feature definition model for the sensor.

19. The computer program product of claim 17 , wherein generating a first feature extraction model of the plurality of feature models that is associated with a first sensor data object of the plurality of sensor data objects comprises:

identifying a generic model and a local model, wherein the local model is generated based at least in part on local training data associated with a physical environment of the first sensor data object;

determining whether a local model confidence measure for the local model exceeds a model confidence threshold value;

responsive to determining that the local model confidence measure for the local model exceeds the model confidence threshold value, adopting the local model as the first feature extraction model; and

responsive to determining that the local model confidence measure for the local model fails to exceed the model confidence threshold value, adopting the generic model as the first feature extraction model.

20. The computer program product of claim 17 , wherein generating a first feature extraction model of the plurality of feature models comprises:

identifying one or more model generating devices associated with the first feature extraction model;

receiving, from each model generating device of the one or more model generating devices, weight update data for the first feature extraction model; and

determining the first feature extraction model based at least in part on each weight update data received from a model generating device of the one or more model generating devices.