Signature use for efficient data transfer

1. A system for identifying data, comprising:

an interface configured to:

receive training data samples associated with interesting events of an interesting event type;

a processor configured to:

train a model to recognize the interesting events;

determine embedded vectors associated with the interesting events; and

determine a representative embedded vector associated with the interesting event type;

a vehicle event recorder interface configured to:

receive sensor data, wherein the sensor data comprises one or more of the following: video data, forward facing video data, inward facing video data, rearward facing video data, side facing video data, audio data, acceleration data, GPS data, LIDAR data, and/or SONAR data; and

receive the representative embedded vector and model data; and

a vehicle event recorder processor configured to:

create the model using the model data;

execute the model to generate a sensor data associated embedded vector; and

determine whether the sensor data is associated with the interesting event type based at least in part on the sensor data associated embedded vector, comprising to:

determine a vector distance between the sensor data associated embedded vector and the representative embedded vector using a machine learning model;

compare the vector distance with a vector distance threshold; and

in response to a determination that the vector distance is less than or equal to the vector distance threshold, determine that the sensor data is associated with the interesting event type.

2. The system of claim 1 , wherein the embedded vectors comprise intermediate outputs of the model.

3. The system of claim 1 , wherein determining a representative embedded vector comprises determining an average of the embedded vectors, determining a weighted average of the embedded vectors, or determining a result of a clustering operation on the embedded vectors.

4. The system of claim 1 wherein determining whether the sensor data is associated with the interesting event type based at least in part on the sensor data associated embedded vector comprises analyzing the sensor data associated embedded vector using a neural network trained on the representative embedded vector.

5. The system of claim 1 wherein executing the model comprises executing only a portion of the model.

6. The system of claim 5 , wherein the portion of the model comprises a first layer or a first set of layers.

7. The system of claim 1 wherein the vehicle event recorder processor is further configured to:

in response to determining that the sensor data is associated with the interesting event type:

provide an indication to upload the sensor data.

8. The system of claim 1 wherein the vehicle event recorder processor is further configured to:

in response to determining that the sensor data associated embedded vector is associated with the interesting event type:

provide an indication to change a storage sample rate for the sensor data.

9. The system of claim 1 wherein the vehicle event recorder processor is further configured to:

in response to determining that the sensor data associated embedded vector is associated with the interesting event type:

provide an indication to change a recording sample rate for the sensor data.

10. The system of claim 1 wherein the vehicle event recorder processor is further configured to:

in response to determining that the sensor data associated embedded vector is associated with the interesting event type:

provide an indication to change a stream type for the sensor data.

11. The system of claim 10 , wherein changing the stream type comprises adding audio to a video stream, adding video to an audio stream, adding accelerometer data to a GPS data stream, increasing a stream data rate, or decreasing the stream data rate.

12. The system of claim 1 wherein the vehicle event recorder processor is further configured to:

in response to determining that the sensor data associated embedded vector is associated with the interesting event type:

provide associated data.

13. The system of claim 12 , wherein associated data comprises higher data rate data, other sensor data, sensor data from a different time, sensor data from a different location, or sensor data from a different vehicle event recorder.

14. A method for identifying data, comprising:

receiving training data samples associated with interesting events of an interesting event type;

training, using a processor, a model to recognize the interesting events;

determining embedded vectors associated with the interesting events; determining a representative embedded vector associated with the interesting event type;

receiving sensor data, wherein the sensor data comprises one or more of the following:

video data, forward facing video data, inward facing video data, rearward facing video data, side facing video data, audio data, acceleration data, GPS data, LIDAR data, and/or SONAR data;

receiving the representative embedded vector and model data;

creating the model using the model data;

executing the model to generate a sensor data associated embedded vector;

determining whether the sensor data is associated with the interesting event type based at least in part on the sensor data associated embedded vector, comprising:

determining a vector distance between the sensor data associated embedded vector and the representative embedded vector using a machine learning model;

comparing the vector distance with a vector distance threshold; and

in response to a determination that the vector distance is less than or equal to the vector distance threshold, determining that the sensor data is associated with the interesting event type.

15. A computer program product for identifying data, the computer program product being embodied in a non-transitory computer readable storage medium and comprising computer instructions for:

receiving training data samples associated with interesting events of an interesting event type;

training a model to recognize the interesting events;

determining embedded vectors associated with the interesting events;

determining a representative embedded vector associated with the interesting event type;

receiving sensor data, wherein the sensor data comprises one or more of the following: video data, forward facing video data, inward facing video data, rearward facing video data, side facing video data, audio data, acceleration data, GPS data, LIDAR data, and/or SONAR data;

receiving the representative embedded vector and model data;

creating the model using the model data;

executing the model to generate a sensor data associated embedded vector;

determining whether the sensor data is associated with the interesting event type based at least in part on the sensor data associated embedded vector, comprising:

determining a vector distance between the sensor data associated embedded vector and the representative embedded vector using a machine learning model;

comparing the vector distance with a vector distance threshold; and

in response to a determination that the vector distance is less than or equal to the vector distance threshold, determining that the sensor data is associated with the interesting event type.