Toward Visualization of EEG Signatures for Anomaly Detection in Dogs Using CEBRA

The cognitive and social abilities of dogs have attracted increasing scientific interest in recent years; however, the neurophysiological mechanisms underlying their social cognition remain underexplored. Event-related potentials (ERPs) have long been used in human research to study attention, expectation, and anomaly detection, often in tasks that involve mismatches between auditory and visual stimuli. Yet ERP studies in dogs are scarce, and the lack of established waveform standards—together with the increased variability that arises when multiple sensory modalities are presented—makes it difficult to obtain labeled data that reflect a dog’s perception of anomalies. In this work, we aim to move toward visualization of EEG signatures for anomaly detection in dogs without relying on supervised learning. Given CEBRA’s capability to identify latent structures in neural data, we use it to embed EEG signals and examine their geometry through visualization. Our preliminary analysis suggests that the embedded representations form distinctive submanifolds around the time when ERPs, potentially reflecting mismatched audiovisual events, are expected to occur. These findings indicate that CEBRA-based visualization can provide a strong foundation for future methods to investigate how dogs process anomalies, offering a potential pathway toward a deeper understanding of canine social cognition.