EgoSelf AI System Uses Graph Memory for Personalized Egocentric Assistance

EgoSelf, a novel AI system, tackles the issue of customizing egocentric assistants by utilizing a graph-based interaction memory that incorporates long-term user data. This memory effectively captures both temporal and semantic connections between interaction events and entities, enabling the creation of user-specific profiles. Personalization is approached as a prediction challenge, where the model anticipates future interactions based on an individual’s historical behavior documented in the graph. Comprehensive experiments validate EgoSelf’s capability to provide meaningful assistance by understanding unique user habits, preferences, and routines. This research was shared on arXiv with the identifier 2604.19564v1, highlighting the ongoing struggle to leverage first-person view data for effective personalization. The graph memory is built from previous observations and includes a specialized learning task for personalization.