Researchers from Mohammed Ayalew Belay, Amirshayan Haghipour, and Pierluigi Salvo Rossi have developed Multi-Episode Prototypical Networks (MEPN) for robust sensor fault diagnosis. Unlike standard prototypical networks, MEPN aggregates prototypes from multiple disjoint support episodes and uses their mean as the final class representative. This approach reduces prototype variance without changing the encoder architecture. The team evaluated MEPN on the DeFACTO sensor dataset using five-way fault classification with synthetic faults. Over 100 independent runs, MEPN reached 98.5% accuracy in the per-episode one-shot setting, significantly outperforming single-episode baselines. Under an equal 10-sample support budget, MEPN and ProtoNet at K=10 were statistically indistinguishable, confirming that prototype accumulation is the key mechanism.
Source: https://arxiv.org/abs/2609.12287