Variance-Regularised Pruning Improves Affective Computing Efficiency
A new pruning framework called Variance-Regularised Pruning (VR) has been introduced to improve computational efficiency in affective computing systems. Unlike existing methods that optimize solely for sparsity, VR explicitly incorporates cross-participant stability into the sparsification process. It evaluates each connection based on its joint contribution to prediction accuracy and variability across users, prioritizing parameters that remain reliable under distributional differences. The approach was evaluated on an unspecified dataset, with results suggesting improved robustness across individuals. This work addresses the need for balancing computational efficiency with reliability in resource-constrained platforms such as adaptive games and assistive technologies.
Key facts
- Variance-Regularised Pruning (VR) is a new pruning framework for affective computing.
- VR incorporates cross-participant stability into the sparsification process.
- Existing pruning methods optimize for sparsity alone without considering robustness across individuals.
- VR evaluates connections based on joint contribution to prediction accuracy and variability across users.
- The approach prioritizes parameters that remain reliable under distributional differences.
- The evaluation was conducted on an unspecified dataset.
- Affective computing systems are increasingly used in adaptive games, assistive technologies, and resource-constrained platforms.
- The work aims to balance computational efficiency with reliability across diverse users.
Entities
—