Systems and methods for data security in collaborative machine learning of autonomous vehicles
View Patent ↗Embodiments are disclosed for a group security scheme for an autonomous vehicle engaged in a collaborative machine learning approach. As an example, a method comprises: generating, in a vehicle, a digital signature based on a first key and a second key, both of the first key and the second key received from a group manager, and transmitting a message signed with the digital signature to a collaborator, the message including coefficients of a local machine learning model of the vehicle. In this way, an accuracy of the local machine learning model of the vehicle may be increased, while a privacy of the vehicle is increased.
1 . A method, comprising:
generating, in a vehicle, a digital signature based on a first key and a second key, both of the first key and the second key received from a group manager, and transmitting a message signed with the digital signature to a collaborator, the message including a subset of coefficients of a local machine learning model of the vehicle, wherein the subset corresponds to a vehicle system functionality; and
receiving an updated machine learning model from the collaborator and adjusting the local machine learning model based on the updated machine learning model, wherein:
adjusting the local machine learning model based on the updated machine learning model includes adjusting at least one coefficient of the local machine learning model;
the updated machine learning model is updated by the collaborator based on data from the vehicle and a plurality of collaborating vehicles of a collaborative machine learning network, wherein each of the plurality of collaborating vehicles uploads a plurality of local machine learning models corresponding to a plurality of vehicle systems to a cloud server, wherein the cloud server analyzes the uploaded machine learning models from the plurality of vehicles, and constructs updated machine learning models, wherein, based on coefficients from each of the local machine learning models, the cloud server determines an updated global machine learning model for each vehicle system, which incorporates learning from each local machine learning model, so that data from each of the plurality of vehicles is incorporated into the global machine learning models, wherein the updated global machine learning model is shared with each of the plurality of vehicles so that the updated global machine learning model replaces local machine learning models, and wherein operation continues, where each of the vehicles continues to train and update their respective local machine learning models, providing continuous, collaborative updates to the machine learning models for each vehicle; and
the first key is a group public key, the group public key distributed to each of the plurality of collaborating vehicles and the collaborator by the group manager, and wherein an unauthorized vehicle that participates in the network has limited ability to provide inaccurate data to the cloud server to data regarding other vehicles in the network by the method removing the unauthorized vehicle by a revocation authority in a location remote from each of the plurality of vehicles.
2 . The method of claim 1 , wherein the second key is a vehicle secret key, the vehicle secret key distinct to the vehicle and distributed to the vehicle by the group manager, and wherein the unauthorized vehicle signs a message using a digital signature generated with another vehicle's global security kit.
3 . The method of claim 2 , wherein the group manager is a computing system communicatively coupled to each of the vehicle, the plurality of collaborating vehicles, and the collaborator.
4 . The method of claim 1 , wherein the local machine learning model of the vehicle is configured to at least partially control operation of a vehicle system of the vehicle without an input from a user of the vehicle.
5 . The method of claim 4 , wherein the vehicle system is one of a steering system, a braking system, an acceleration system, a transmission system, an object detection system, a cruise control system, and a climate control system.
6 . A method, comprising:
receiving, at a vehicle, a vehicle secret key and a group public key from a group manager;
adjusting, at the vehicle, coefficients of a local machine learning model of the vehicle based on sensor data and connected autonomous vehicle sub-systems outputs from at least one sensor of the vehicle;
generating, at the vehicle, a digital signature based on the vehicle secret key and the group public key;
transmitting coefficients of the local machine learning model corresponding to a vehicle functionality to a collaborator via a message from the vehicle signed with the digital signature;
receiving, at the vehicle, coefficients of an updated machine learning model from the collaborator; and
adjusting a vehicle system of the vehicle based on the updated machine learning model, wherein the group public key is distributed to each of a plurality of collaborating vehicles of a collaborative machine learning network and the collaborator by the group manager, wherein each of the plurality of collaborating vehicles uploads a plurality of local machine learning models corresponding to a plurality of vehicle systems to a cloud server, and wherein an unauthorized vehicle that participates in the network has limited ability to provide inaccurate data to the cloud server to data regarding other vehicles in the network by the method removing the unauthorized vehicle by a revocation authority in a location remote from each of the plurality of collaborating vehicles.
7 . The method of claim 6 , wherein each of the group manager and the collaborator are processors external to the vehicle, each of the group manager and the collaborator communicatively coupled to the vehicle, and the group manager communicatively coupled to the collaborator.
8 . The method of claim 6 , wherein the updated machine learning model from the collaborator is based in part on the coefficients of the local machine learning model.
9 . The method of claim 6 , wherein the local machine learning model is a machine learning algorithm trained to autonomously control the vehicle system of the vehicle.
10 . The method of claim 6 , further comprising:
receiving, at the vehicle, a second vehicle secret key and a second group public key from the group manager;
adjusting, at the vehicle, coefficients of a second local machine learning model of the vehicle based on sensor data from at least one sensor of the vehicle;
generating, at the vehicle, a second digital signature based on the second vehicle secret key and the second group public key;
transmitting the coefficients of the second local machine learning model to the collaborator via a message from the vehicle signed with the second digital signature;
receiving, at the vehicle, coefficients of a second updated machine learning model from the collaborator; and
adjusting second vehicle system of the vehicle based on the second updated machine learning model.
11 . A system, comprising:
a vehicle system of a vehicle;
an autonomous vehicle perception, path planning, and control system including a local machine learning model, the local machine learning model corresponding to the vehicle system and configured to at least partially control the vehicle system; and
a controller storing executable instructions in non-transitory memory that, when executed, cause the controller to:
receive a group public key and a vehicle secret key from a remote server via a wireless connection;
update coefficients of the local machine learning model based on data from at least one sensor of the vehicle;
generate a digital signature based on the group public key and the vehicle secret key;
transmit a message signed with the digital signature to the remote server, the message including coefficients of the local machine learning model updated based on the data from the at least one sensor during vehicle operation;
receive an updated machine learning model from the remote server, wherein the updated machine learning model is updated by the remote server based on data from the vehicle and a plurality of collaborating vehicles of a collaborative machine learning network, wherein each of the collaborating vehicles uploads a plurality of local machine learning models corresponding to a plurality of vehicle systems to a cloud server; and
adjust the local machine learning model based on the updated machine learning model, wherein adjusting the local machine learning model based on the updated machine learning model includes adjusting at least one coefficient of the local machine learning model, wherein the group public key is distributed to each of the plurality of collaborating vehicles by the remote server, and wherein an unauthorized vehicle that participates in the network has limited ability to provide inaccurate data to the cloud server to data regarding other vehicles in the network by the method removing the unauthorized vehicle by a revocation authority in a location remote from each of the plurality of vehicles.
12 . The system of claim 11 , wherein the remote server includes a group manager, a collaborator, and a revocation authority.
13 . The system of claim 11 , wherein the vehicle system is one of a braking system, an acceleration system, a steering system, a cruise control system, a climate control system, an object detection system, a lighting system, and a transmission system.
14 . The system of claim 11 , wherein the digital signature is generated based on one of a BBS group signature scheme, an ACJT group signature scheme, a CG group signature scheme, BS group signature scheme, and a group signature scheme with dynamic registration.
15 . The system of claim 11 , wherein the autonomous vehicle perception, path planning, and control system at least partially controls vehicle operation with an input from a user.