17818967. HIDDEN MACHINE LEARNING FOR FEDERATED LEARNING simplified abstract (Capital One Services, LLC)
Contents
HIDDEN MACHINE LEARNING FOR FEDERATED LEARNING
Organization Name
Inventor(s)
Jeremy Goodsitt of Champaign IL (US)
HIDDEN MACHINE LEARNING FOR FEDERATED LEARNING - A simplified explanation of the abstract
This abstract first appeared for US patent application 17818967 titled 'HIDDEN MACHINE LEARNING FOR FEDERATED LEARNING
Simplified Explanation
The method involves providing a distributed instance of a machine learning model to a client computing device and access to a predetermined token associated with a predetermined label to a client-side application at the client computing device. The client-side application detects a user input containing the predetermined token, sends the user input to the distributed instance to get a predicted label, and updates the model parameters based on the predicted label and the predetermined label. The machine learning model is updated by obtaining the updated model parameters from the client computing device.
- Distributed machine learning model provided to client computing device
- Client-side application detects user input with predetermined token
- Predicted label obtained from distributed instance
- Model parameters updated based on predicted label and predetermined label
- Machine learning model updated with updated model parameters from client computing device
Potential Applications
- Personalized recommendations
- Fraud detection
- Sentiment analysis
Problems Solved
- Efficient updating of machine learning models
- Real-time prediction and updating based on user input
Benefits
- Improved accuracy of machine learning models
- Faster response time to user input
- Reduced computational load on central servers
Original Abstract Submitted
A method includes providing a distributed instance of a machine learning model to a client computing device and access to a predetermined token associated with a predetermined label to a client-side application at the client computing device. The client-side application is configured to cause the client computing device to detect a user input that includes the first predetermined token, provide the user input to the distributed instance to obtain a first predicted label for the first predetermined token, and update a set of model parameters of the distributed instance based on the first predicted label and the first predetermined label. The method may also include updating the machine learning model by obtaining the one or more updated model parameters of the distributed instance from the client computing device.
