Reviews: EEG-GRAPH: A Factor-Graph-Based Model for Capturing Spatial, Temporal, and Observational Relationships in Electroencephalograms

SUMMARY: The authors propose a probabilistic model and MAP inference for localizing seizure onset zones (SOZ) using intracranial EEG data. The proposed model captures spatial correlations across EEG channels as well as temporal correlations within a channel. The authors claim that modeling these correlations leads to improved predictions when compared to detection methods that ignore temporal and spatial dependency. PROS: This is a fairly solid applications paper, well-written, well-motivated, and an interesting application. CONS: The proof of Prop. 1 is not totally clear, for example the energy in Eq. (4) includes a penalty for label disagreement across channels, which is absent in the the graph cut energy provided by the proof.