Verifying sensor data using embeddings

Patent metadata

Field	Value
Publication number	US-2018310135-A1
Application number	US-201715495686-A
Country	US
Kind code	A1
Filing date	Apr 24, 2017
Priority date	Apr 24, 2017
Publication date	Oct 25, 2018
Grant date	—

Abstract

A network system analyzes data samples using embeddings based on, for example, symbolic representations of the data samples or representations in latent dimension space. The network system coordinates providers who provide geographical location-based services to users. The network system may receive data samples from the client device of a provider. For instance, a sensor of the client device captures the data samples during a transportation service along a particular route. To verify that the data samples accurately indicate the location or movement of the provider, the network system can generate a test embedding representing the data samples and compare the test embedding with a reference embedding. The reference embedding is generated based on data samples captured for other similar services, e.g., corresponding to providers who also provided transportation services along the same particular route.

First claim

1 . A method comprising:
receiving, from a client device, data associated with sensor information of the client device and associated with a trip record, the data including (i) a data sample for a set of sensors of the client device and (ii) one or more characteristics of the data sample; generating a test embedding for the data sample, the test embedding using a plurality of latent dimensions that represent at least a portion of the data sample; identifying a reference embedding for a set of reference characteristics, the set of reference characteristics corresponding to at least one of the one or more characteristics of the data sample, the reference embedding being based on a set of embeddings each using the plurality of latent dimensions that represent sensor data for a set of trip records different than the trip record, the set of trip records being associated with the set of reference characteristics; determining a similarity score between the test embedding corresponding to the trip record and the reference embedding by comparing each latent dimension of the test embedding and a corresponding latent dimension of the reference embedding; and verifying, in response to the similarity score exceeding a threshold score, that the data sample was captured while the set of sensors were subject to the set of reference characteristics.

2 . The method of claim 1 , wherein the test embedding and the reference embedding are generated using a model trained based at least in part on feature vectors derived from data samples captured for the set of trip records.

3 . The method of claim 1 , wherein the portion of the data sample has a first duration in time, and wherein generating the test embedding comprises:
generating a plurality of embeddings for a plurality of sub-portions of the portion of the data sample, each of the sub-portions having a second duration in time less than the first duration in time; and aggregating the plurality of embeddings.

4 . The method of claim 1 , wherein a first user is associated with the client device and the trip record, wherein the set of trip records includes at least a sample trip taken by a second user, and wherein the set of embeddings includes a sample embedding representing sample sensor data captured for the sample trip by another set of sensors of another client device of the second user.

5 . The method of claim 4 , wherein the sample trip includes a plurality of routes, and further comprising:
determining a route of the plurality of routes traveled by both the first user and the second user based at least in part on the similarity score.

6 . The method of claim 1 , wherein the set of trip records includes at least the trip record, and wherein the set of embeddings includes a sample embedding representing sample sensor data captured for the trip record by another sensor of the client device not included in the set of sensors.

7 . The method of claim 1 , wherein the one or more characteristics of the data sample includes at least one of: an origin or destination location of the trip record, a route of the trip record, a type of the client device, or a user of the client device.

8 . The method of claim 1 , further comprising:
determining that the one or more characteristics of the data sample describe a geophysical event; determining, for the trip record, a likelihood score that the geophysical event occurred based at least in part on the similarity score, and wherein the set of reference characteristics is associated with the geophysical event.

9 . The method of claim 1 , wherein the client device is transported in a vehicle, and wherein the method further comprises:
determining that the one or more characteristics of the data sample describe a safety incident; determining that the vehicle was involved in the safety incident based at least in part on the similarity score.

10 . A method comprising:
receiving, from a client device of a first user, data associated with sensor information of the client device and associated with a trip record, the data including (i) a data sample for a set of sensors of the client device and (ii) one or more characteristics of the data sample; generating a test embedding for the data sample, the test embedding using a plurality of latent dimensions that represent at least a portion of the data sample; identifying a reference embedding for a set of reference characteristics, the set of reference characteristics corresponding to at least one of the one or more characteristics of the data sample, the reference embedding being based on a set of embeddings each using the plurality of latent dimensions that represent sensor data for a set of trip records different than the trip record, the set of trip records being associated with the set of reference characteristics, the reference embedding generated using a model trained based at least in part on feature vectors derived from data samples captured for the set of trip records; determining a similarity score between the test embedding corresponding to the trip record and the reference embedding by comparing each latent dimension of the test embedding and a corresponding latent dimension of the reference embedding; and verifying, in response to the similarity score exceeding a threshold score, that the first user and a second user both traveled along one or more routes associated with the trip record.

11 . The method of claim 10 , further comprising:
determining a route of the one or more routes along which only the first user or the second user traveled for the trip record.

12 . A computer program product comprising a non-transitory computer readable storage medium having instructions encoded thereon that, when executed by one or more processors, cause the one or more processors to:
receive, from a client device, data associated with sensor information of the client device and associated with a trip record, the data including (i) a data sample for a set of sensors of the client device and (ii) one or more characteristics of the data sample; generate a test embedding for the data sample, the test embedding using a plurality of latent dimensions that represent at least a portion of the data sample; identify a reference embedding for a set of reference characteristics, the set of reference characteristics corresponding to at least one of the one or more characteristics of the data sample, the reference embedding being based on a set of embeddings each using the plurality of latent dimensions that represent sensor data for a set of trip records different than the trip record, the set of trip records being associated with the set of reference characteristics; determine a similarity score between the test embedding corresponding to the trip record and the reference embedding by comparing each latent dimension of the test embedding and a corresponding latent dimension of the reference embedding; and verify, in response to the similarity score exceeding a threshold score, that the data sample was captured while the set of sensors were subject to the set of reference characteristics.

13 . The non-transitory computer readable storage medium of claim 12 , wherein the test embedding and the reference embedding are generated using a model trained based at least in part on feature vectors derived from data samples captured for the set of trip records.

14 . The non-transitory computer readable storage medium of claim 12 , wherein the portion of the data sample has a first duration in time, and wherein generating the test embedding comprises:
generating a plurality of embeddings for a plurality of sub-portions of

Assignees

• Uber Technologies Inc

Inventors

• Cirit Fahrettin Olcay

Classifications

• H04W4/029 Primary

  Location-based management or tracking services · CPC title

• H04W4/023

  using mutual or relative location information between multiple location based services [LBS] targets or of distance thresholds · CPC title

• H04L67/12

  specially adapted for proprietary or special-purpose networking environments, e.g. medical networks, sensor networks, networks in vehicles or remote metering networks · CPC title

• G08G1/205

  Indicating the location of the monitored vehicles as destination, e.g. accidents, stolen, rental · CPC title

• H04W4/40

  for vehicles, e.g. vehicle-to-pedestrians [V2P] · CPC title