Adaptive Hybrid Sort for Multimodal Edge Reasoning in Resource-Constrained Environments

Shrinivass Arunachalam Balasubramanian

Abstract: We introduce Adaptive Hybrid Sort, a data-aware preprocessing paradigm for multimodal AI systems deployed at the edge in healthcare, robotics, and safety-critical settings. Using entropy-driven strategy selection, our framework dynamically chooses among Counting Sort, Radix Sort, and QuickSort based on input distributions. A feature extraction module coupled with a FSM-guided classifier enables efficient sorting tailored to modality-specific sensor feeds. Evaluations on synthetic and real-world datasets show marked improvements in execution time, adaptability, and resource utilization, positioning this approach as a scalable solution for domain-informed multimodal reasoning pipelines.