Dongyang Kuang, Zizheng Ma, Yushan Zhang +1 more
Most EEG emotion classifiers treat feelings as unrelated boxes, so mistaking calm for excited counts the same as mistaking calm for content, even though one error is far more absurd than the other. This paper bakes in the psychology instead. It arranges emotions as nodes on a graph, with edges set by dimensional emotion theory so that similar feelings sit close together, then penalizes the model when its predictions ignore that layout.
They try three ways to enforce it, from simple soft labeling up to an optimal-transport distance, and test all three across three different network backbones. The payoff is modest but consistent: up to about 5.4% better accuracy, and, more interestingly, a 39% cut in the kind of misclassifications that make no psychological sense.
Figures are from the abstract, so read the paper for per-dataset results and how the emotion graph was built.
EEG-based emotion recognition is critical for mental health monitoring and affective brain-computer interfaces, yet existing deep learning approaches often treat emotion classes as isolated labels, ignoring their psychological interdependencies. We propose a graph-regularized learning framework that conceptualizes emotions as nodes in a graph where edges encode proximity based on dimensional emotion theories. We adapt three complementary regularization strategies--Graph…
Event-based Neural Decoding for Neuroprosthetic Motor Control
arXiv (EEG) · July 13, 2026Graph Bispectrum for Nonlinear Mode Interactions
arXiv (BCI) · July 12, 2026Threat Vectors and the State of the Art in Defense Methods for Security in Neurotechnology
arXiv (EEG) · July 10, 2026PHINN-EEG: Topological Time-Series Analysis of Dream-State EEG -- Dynamic Betti Curves for Dream Content Classification and Topology-Conditioned Neural Signal Synthesis