Resource-Constrained Affect Modelling via Variance Regularisation Pruning

2026-05-26

AI Summary

This work addresses a critical limitation in existing model pruning methods for resource-constrained affective computing: their exclusive focus on sparsity while neglecting the impact of parameter removal on cross-user performance stability. To remedy this, the authors propose Variance-Regularized (VR) pruning, a novel framework that explicitly incorporates cross-user robustness into the pruning criterion. By jointly optimizing prediction accuracy and inter-subject performance variance, VR prioritizes retaining parameters that are robust to distributional shifts across users. The method employs a connection-sensitivity-based pruning strategy and is evaluated using the Concordance Correlation Coefficient (CCC) on the AGAIN dataset. Experimental results demonstrate that, even at an 80% pruning ratio without fine-tuning, the approach maintains competitive CCC performance, significantly enhancing the generalization capability and deployment feasibility of lightweight affective models in real-world interactive scenarios.

Abstract

Affective computing systems are increasingly embedded in pervasive and interactive environments, such as adaptive games, assistive technologies, and resource-constrained platforms, where computational efficiency must be balanced with reliability across diverse users. Model pruning offers an effective way to reduce computational demands, yet existing approaches typically optimise for sparsity alone, without accounting for how parameter removal impacts robustness across individuals. In this work, we introduce Variance-Regularised Pruning (VR), a pruning framework that explicitly incorporates cross-participant stability into the sparsification process. Rather than relying solely on average prediction error, VR evaluates each connection based on its joint contribution to both prediction accuracy and variability across users, prioritising parameters that remain reliable under distributional differences. We evaluate the proposed approach on the AGAIN dataset, which includes arousal annotations collected across nine affect-eliciting game environments. Experimental results demonstrate that VR maintains competitive Concordance Correlation Coefficient (CCC) performance even at 80\% sparsity without additional fine-tuning, highlighting its suitability for deployment in real-world, resource-limited affect-aware systems. Overall, the proposed framework supports the development of compact, robust affective models that can operate reliably in real-world interactive environments.

Problem

• affective computing
• model pruning
• cross-user robustness
• resource-constrained systems
• variance regularisation

Innovation

• Variance-Regularised Pruning
• affective computing
• model pruning
• cross-participant robustness
• resource-constrained systems