How does TS-GAC ensure spatial consistency during contrastive learning?

TS-GAC ensures spatial consistency by combining graph-based augmentations with graph-level contrasting. It creates weak and strong views through node and edge perturbations, then applies Node-level Contrasting, which enforces similarity between corresponding sensors across views, and Graph-level Contrasting, which aligns global features of the same sample across views while separating different samples.

TS-GAC builds its contrastive learning framework around spatial consistency. During augmentation, node frequency augmentations and temporal augmentations within segments create weak and strong views, and edge augmentations perturb sensor correlations. The resulting graphs are processed by a GNN-based encoder. Spatial consistency is then enforced by graph contrasting, which includes Node-level Contrasting (NC) and Graph-level Contrasting (GC). NC pushes the representations of corresponding sensors to be similar across the two views of the same sample, making sensor-level features robust. GC aligns the global graph-level features of the same sample across views and differentiates them from the global features of other samples in the batch.