Resource-Constrained Affect Modelling via Variance Regularisation Pruning

In three linesVariance-Regularised Pruning (VR) is a model compression method for affective computing that prioritises cross-user stability over sparsity alone. Tested on the AGAIN dataset (9 game environments), VR maintains competitive CCC performance at 80% sparsity without additional fine-tuning, suited for resource-constrained embedded systems.

Summary generated by Claude — human-verified