Memory Bank
A core component of a classic class of industrial anomaly detection methods. During training, features are extracted from normal samples and stored as an explicit feature set; at inference, nearest-neighbor retrieval determines whether a test feature deviates from that set.
Timeline (last 90 days)
Memory bank methods (such as PatchCore and PaDiM) store a set of normal sample features during training and determine anomalies at inference via nearest-neighbor retrieval, with storage cost growing linearly with the training set size.
In contrastive learning (such as MoCo), the memory bank is used to store features of a large number of negative samples so as to compute a more stable contrastive loss under limited batch size; SimOPD adapts it into a paired form that simultaneously stores student-generated response sequences and their corresponding teacher internal representations.
All Facts (2)
Memory bank methods (such as PatchCore and PaDiM) store a set of normal sample features during training and determine anomalies at inference via nearest-neighbor retrieval, with storage cost growing linearly with the training set size.
50%UnverifiedIn contrastive learning (such as MoCo), the memory bank is used to store features of a large number of negative samples so as to compute a more stable contrastive loss under limited batch size; SimOPD adapts it into a paired form that simultaneously stores student-generated response sequences and their corresponding teacher internal representations.
50%