On-Device Model for Smart Replies
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing devices, especially mobile ones like smartphones and smartwatches, face challenges in providing efficient and power-conserving ways to respond to communications due to limited resources and the need for reliable, quick, and contextually relevant responses, especially in situations where typing or speaking is cumbersome or socially unacceptable.
Innovation Solution
Implementing an on-device machine-trained model that generates candidate responses based on user input, utilizing machine learning techniques to predict relevant replies without relying on external models or networks, thus enabling quick and inconspicuous responses while conserving battery life and ensuring end-to-end encryption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If an on-device machine-trained model is implemented, then response generation speed and battery conservation are improved, but device complexity increases
Solution Approach 1:
The machine-trained model is trained offline in advance and deployed to the device, so that during actual use the model can generate responses quickly without requiring real-time training or complex online computation, thus achieving fast response generation while keeping runtime device complexity manageable
Solution Approach 2:
Instead of implementing complex real-time natural language processing capabilities on the device, a pre-trained model (a copy of the intelligence) is deployed to the device, enabling fast local response generation without requiring the full complexity of training systems to be present during operation
2Use of energy by moving object
If local processing is used, then power consumption is reduced and security is improved, but response accuracy may worsen
Solution Approach 1:
The model is optimized for mobile devices by adjusting parameters such as model size, complexity, and computation depth to suit the constraints of mobile hardware, enabling accurate enough responses to be generated locally with acceptable power consumption while maintaining sufficient accuracy for the application context
3Adaptability or versatility
If on-device model is used, then network dependency is reduced, but model size and memory requirements increase
Solution Approach 1:
The essential response generation capability is extracted from the cloud and placed locally on the device in the form of a compact model, enabling the device to function independently of network connectivity while the model size is kept manageable through optimization techniques
Data Source
AI summary
A computing device may receive a communication sent from an external computing device. At least one processor of the computing device may determine, using an on-device machine-trained model and based at least in part on the communication, one or more candidate responses to the communication. The at least one processor may receive an indication of a user input that selects a candidate response from the one or more candidate responses. Responsive to receiving the indication of the user input that selects the candidate response, the at least one processor may send the candidate response to the external computing device.