Hyper-MDR: An open-world multimodal reasoning framework based on dynamic hypergraph and meta-strategy optimization.

This work proposes a new paradigm for open-world multimodal perception that integrates semantic depth and adaptability, and proposes a meta-policy gradient-based adaptive controller, which dynamically adjusts feature fusion weights, propagation depth, and attention topology based on the detection state and historical trajectories.