IMS AI Model Signalling for Partial Inference on Lightweight UEs

The AI/ML model is divided into multiple segments or components that can be selectively delivered and executed. The patent describes delivering model parameters, weights, and configuration data separately from the inference engine, allowing lightweight devices to receive only necessary model portions while heavier processing is performed on the network side or more capable devices.

The patent introduces an intermediary mechanism (SDP signalling and model delivery framework) between the model provider and the UE device. This intermediary manages the complex model delivery process, handles compatibility between different devices and models, and optimizes resource allocation, reducing the burden on individual UE devices.

The patent extracts and delivers only the essential model components (parameters, weights, biases) that are necessary for inference, rather than transmitting complete model files. The SDP signalling mechanism selectively extracts and delivers model elements based on device capabilities and service requirements, reducing unnecessary data transmission while maintaining inference accuracy.

The patent changes the representation format of model data from complete model structures to compressed parameter sets. By transmitting model parameters, weights, and configuration data in optimized formats through SDP signalling, the system reduces data size while preserving the functional accuracy of the AI/ML models.

The patent implements a universal SDP signalling framework that can handle multiple AI/ML model types, formats, and delivery scenarios through a single standardized interface. This universal mechanism works across different UE devices, network configurations, and model providers, ensuring compatibility without requiring device-specific customization logic.

Instead of having each device implement complex compatibility logic for different model formats, the patent inverts the approach by having the model delivery system adapt to device capabilities. The SDP signalling mechanism receives capability information from devices and automatically configures appropriate model delivery parameters, reversing the complexity burden from the device side to the network side.