Abstract
Intelligent agents on shared computing devices, such as smart displays or in-vehicle infotainment systems, may not effectively serve groups of concurrent users, as they can be tied to a single user account or a generic profile, which can limit their ability to adapt to a specific group composition. After obtaining appropriate permissions and consent, the disclosed technology can use on-device sensors, for example, cameras and microphones, to detect and identify multiple users. For each identified user, a corresponding personalized, parameter-efficient adaptation module can be retrieved. These modules may then be dynamically weighted based on real-time contextual signals, such as user proximity or active speaker status, and synthesized to create a transient, group-optimized model instance. This approach can enable a generative model to adapt its behavior to the collective context of the group, potentially improving the relevance and safety of the shared user experience without requiring manual profile changes.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Kushwaha, Vivek, "Sensor-Driven Synthesis of Group-Optimized Models Using Personalized Adapters", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12086