IoT Uplink Scheduling via Contributiveness Metrics

The patent applies local quality by differentiating device contributiveness based on local conditions such as observation positions and path losses. Each device is evaluated individually using a contributiveness metric that reflects its specific ability to provide useful information for downstream tasks, rather than applying uniform scheduling across all devices. This resolves the contradiction by allocating resources based on local device characteristics and their actual contribution potential.

The patent implements partial action by selecting only a subset of devices for scheduling based on their contributiveness metrics. Instead of scheduling all devices, the system identifies and schedules only those devices that are most likely to provide useful information for downstream tasks. This partial scheduling approach improves system efficiency by concentrating resources on contributive devices while excluding non-contributive ones, directly resolving the contradiction between productivity and quantity of scheduled devices.

The patent applies preliminary action by pre-training the deep neural network to learn device contributiveness metrics before actual scheduling operations. The DNN is trained offline to polarize weights that identify contributive devices, so that during runtime, the scheduling decision can be made quickly based on pre-learned metrics. This preliminary training phase separates the computationally intensive learning process from the real-time scheduling operation, resolving the contradiction between resource allocation efficiency and computational complexity.