00:00

Question 57·Hard·Command of Evidence

Battery use can limit how useful machine-learning apps are on phones. In a study of on-device image recognition, computer scientist Elena Park compared three models: TinyNet, SparseNet, and DenseNet. Park estimated each model’s efficiency by dividing its top-1 accuracy by its battery drain (measured in milliamp-hours), reasoning that a higher accuracy-per-battery value indicates a more efficient model. Park concluded that SparseNet is more battery-efficient than DenseNet.

Which choice best describes data in the graph that support Park’s conclusion?