Brief

Researchers have introduced AdaTKG, a novel approach to temporal knowledge graph reasoning that models entities as adaptive processes. Unlike static methods, AdaTKG refines an entity's representation each time it participates in a fact, storing this information in a per-entity memory. This adaptive memory accumulates over time, improving prediction accuracy as more interactions are observed. A key feature is the use of a learnable exponential moving average for memory updates, allowing AdaTKG to handle unseen entities effectively. AI