These tools and metrics are designed to help AI actors develop and use trustworthy AI systems and applications that respect human rights and are fair, transparent, explainable, robust, secure and safe.
Sparsity
While smoothness and spatial locality capture spatial properties, the individual values shall also be sparse, since few highly important regions are more indicative of a good explanation than several mildly relevant ones. This is why a sparsity metric should be used. Please refer to the reference URL for full formulation.
Trustworthy AI Relevance
This metric addresses Environmental Sustainability and Robustness by quantifying relevant system properties. Environmental Sustainability: Sparsity (measured as percent-zero weights, L0 norm, or fraction of inactive activations) directly correlates with reduced parameter count, fewer FLOPs, lower memory bandwidth and smaller model storage. These reductions translate into reduced energy consumption and carbon footprint during training and inference.
Related use cases :
Adaptive Neural Connections for Sparsity Learning
Uploaded on Nov 1, 2023About the metric
You can click on the links to see the associated metrics
Objective(s):
Purpose(s):
Target sector(s):
Lifecycle stage(s):
Target users:
Partnership on AI
