Get started
Work

Cross-modal attention and Q-learning-based intelligent scheduling for interdisciplinary digital media systems

Year: 2026
Type: article
Abstract: Interdisciplinary digital media platforms receive concurrent image, speech, and text interaction tasks whose semantic urgency, computational demand, and service deadlines vary over time. This study formulates the problem as online scheduling of multimodal interaction tasks across heterogeneous GPU, CPU, and edge-computing nodes under queue-capacity, node-compatibility, and deadline constraints. A cross-modal attention module is used to generate a unified task descriptor and estimate the semantic consistency of the available modalities. The resulting semantic information is introduced into task pre-ranking and system-state construction rather than being treated as an independent recognition output. A tabular Q-learning scheduler then selects among ten fixed scheduling-rule and resource-node combinations, thereby avoiding an action space that changes with the number of pending tasks and available nodes. The experimental benchmark is constructed from public image, speech, and text datasets through an explicit intent-matching protocol, while a constraint-optimization oracle provides reference decisions for evaluating scheduling accuracy. Under the reported high-concurrency setting, the proposed method obtains an average queue-to-dispatch latency of 192 ms and an oracle-consistent scheduling accuracy of 87.2%. Its end-to-end task completion time on the edge device is 1.589 s, indicating suitability for delay-tolerant interactive services rather than hard real-time applications. The results support the feasibility of semantically informed scheduling in educational and cultural digital media services.
DOI: 10.1007/s42452-026-09470-8
Author: Guixia Wang
Language: English
Open Access status: diamond
OpenAlex ID: W7208760269
FWCI:
Cites:
Cited by:
Related to: