Back to results
US 12,603,915 B1Granted
Agentic on-device adaptation for increasing node efficiency by accommodating distribution shifts
AssigneeU.S. Bancorp, National Association
InventorsRahul Sharma, Narasimha ‘Murthy’ M Kunchakarra
Life of this patent
Estimating the term…
Abstract
A method and related systems for dynamically adjusting on-device configurations for models based on detected context switches is disclosed. The system can use an on-device language model to determine context parameters in a first mode and then uses the context parameters to determine prediction models or prediction model parameters to use for inputs provided in a second mode.
Claims20
A system comprising:
- at least one processor;
- and
- memory storing instructions that, when executed by the at least one processor, cause the system to:
- receive first input data via a first data modality on a first electronic device;
- determine, using an on-device language model, at least one first context parameter from the first input data;
- correlate the first context parameter with second context parameters detected from a second, different data modality;
- configure a prediction model for processing subsequent input based on the correlation between the first context parameter and the second context parameters, wherein configuring the prediction model comprises selecting the prediction model from a plurality of pr-diction models stored in an on-device model database by referencing an associative data structure that maps the correlated context parameters to the prediction model, wherein the first context parameter determined from the first data modality is used to select the prediction model for processing input received via the second data modality;
- and
- generate an output by providing further input to the configured prediction model, and present, based on the output, a message or indication via a user interface of the first electronic device.
6 more elements
A method comprising:
- receiving, by a first electronic device, first input via a first data modality;
- determining, using an on-device language model, at least one first context parameter from the first input;
- correlating the first context parameter with a second context parameter detected from data received via a second, different data modality;
- configuring a prediction model based on the correlation between the context parameters, wherein configuring the prediction model comprises selecting the prediction model from a plurality of prediction models stored in an on-device model database by referencing an associative data structure that maps the correlated context parameters to the prediction model, wherein the first context parameter determined from the first data modality is used to select the prediction model for processing input received via the second data modality;
- providing an input to the selected or configured prediction model to obtain an output;
- and
- causing a user interface of the first electronic device to present a message or output based on the output.
4 more elements
A non-transitory computer-readable medium storing instructions that, when executed by at least one processor, cause the processor to perform operations comprising:
- receiving, on an electronic device, a first input via a first data modality;
- determining at least one first context parameter using an on-device language model;
- correlating the first context parameter with a second context parameter derived from a different data modality;
- selecting a prediction model from a plurality of prediction models stored in an on-device model database by referencing an associative data structure that maps the correlated context parameters to the prediction model, wherein the first context parameter determined from the first data modality is used to select the prediction model for processing input received via the second data modality;
- A generating an output with the prediction model;
- and
- presenting, via a user interface, a message representing the output.
4 more elements
Dependencies are inferred from the claim text.