Modular network based knowledge sharing for multiple entities
A method for vehicle fault detection is provided. The method includes training, by a cloud module controlled by a processor device, an entity-shared modular and a shared modular connection controller. The entity-shared modular stores common knowledge for a transfer scope, and is formed from a set of sub-networks which are dynamically assembled for different target entities of a vehicle by the shared modular connection controller. The method further includes training, by an edge module controlled by another processor device, an entity-specific decoder and an entity-specific connection controller. The entity-specific decoder is for filtering entity-specific information from the common knowledge in the entity-shared modular by dynamically assembling the set of sub-networks in a manner decided by the entity specific connection controller.
1 . A method for vehicle fault detection, comprising:
training, by a cloud module controlled by a processor device, an entity-shared modular and a shared modular connection controller, the entity-shared modular storing common knowledge for a transfer scope, the transfer scope being based on a local community where an entity is located, and formed from a set of sub-networks which are dynamically assembled for different target entities of a vehicle by the shared modular connection controller, and the shared modular connection controller connects the sub-networks; and
training, by an edge module controlled by another processor device, an entity-specific decoder, an entity-specific connection controller, and an output module, the entity-specific decoder for filtering entity-specific information from the common knowledge in the entity-shared modular by dynamically assembling the set of sub-networks in a manner decided by the entity specific connection controller, the output module transforms a dimension of the shared decoder to an output dimension to adapt to different task needs, wherein the entity-shared modular and the entity-specific decoder form a task-specific graph using task and data scenarios.
2 . The method of claim 1 , further comprising detecting, by the edge module based on the entity-specific information filtered from the common knowledge, an anomaly in at least one of the different target entities of the vehicle.
3 . The method of claim 1 , further comprising controlling a system of the vehicle for obstacle avoidance responsive to a detection of the anomaly in at least one of the different target entities of the vehicle.
4 . The method of claim 3 , wherein the system of the vehicle is selected from a group consisting of steering, accelerating, and braking.
5 . The method of claim 1 , wherein the entity-shared modular and the shared modular connection controller are jointly trained.
6 . The method of claim 1 , wherein the entity-specific controller and the entity-specific connection controller are trained only with the entity-specific information while excluding the common knowledge.
7 . The method of claim 1 , wherein parameters of the entity-shared modular and the shared modular connection controller are frozen while training the entity-specific connection controller.
8 . The method of claim 1 , wherein the task-specific graph measures a similarity between the different target entities such that connections in the task-specific graph reflect common attributes between the different target entities.
9 . The method of claim 8 , wherein nodes in the task-specific graph represent respective ones of the different target entities, and edges in the task-specific graph represent common attributes between respective pairs of the different target entities.
10 . The method of claim 1 , wherein the shared modular connection controller decides different decision connections between different layers having different ones of the sub-networks in the set.
11 . The method of claim 1 , wherein the shared modular connection controller and the entity-specific connection controller selectively make or break a connection with the entity-shared modular and the entity-specific decoder, respectively.
12 . The method of claim 11 , wherein the shared modular connection controller implements a reinforcement learning process using a reward in relation to making and braking the connection.
13 . The method of claim 1 , wherein the cloud module is pre-trained prior to the edge module.
14 . The method of claim 1 , wherein the entity-shared modular and the entity specific decoder form a personalized model for at least one of the different target entities.
15 . A system for vehicle fault detection, comprising:
a cloud module, controlled by a processor device, for training an entity-shared modular and a shared modular connection controller, the entity-shared modular storing common knowledge for a transfer scope, the transfer scope being based on a local community where an entity is located, and formed from a set of sub-networks which are dynamically assembled for different target entities of a vehicle by the shared modular connection controller, and the shared modular connection controller connects the sub-networks; and
an edge module controlled by another processor device, for training an entity-specific decoder, an entity-specific connection controller, and an output module, the entity-specific decoder for filtering entity-specific information from the common knowledge in the entity-shared modular by dynamically assembling the set of sub-networks in a manner decided by the entity specific connection controller, the output module transforms a dimension of the shared decoder to an output dimension to adapt to different task needs, wherein the entity-shared modular and the entity-specific decoder form a task-specific graph using task and data scenarios.
16 . The system of claim 15 , wherein the edge module detects an anomaly in at least one of the different target entities of the vehicle, based on the entity-specific information filtered from the common knowledge.
17 . The system of claim 15 , wherein the edge module interacts to initiate a control of a system of the vehicle for obstacle avoidance responsive to a detection of the anomaly in at least one of the different target entities of the vehicle.
18 . The system of claim 17 , wherein the system of the vehicle is selected from a group consisting of steering, accelerating, and braking.
19 . The system of claim 15 , wherein the entity-shared modular and the shared modular connection controller are jointly trained.
20 . The system of claim 15 , wherein the entity-specific controller and the entity-specific connection controller are trained only with the entity-specific information while excluding the common knowledge.