iCoTASC framework cuts semantic communication overhead with hybrid offline-online design

Task-oriented semantic communication faces a critical trade-off: allocating scarce radio resources to semantic features under fast-fading wireless conditions and strict latency budgets. Existing approaches are either optimization-heavy (causing prohibitive online overhead) or rely on end-to-end retraining with slowly varying channel assumptions. The paper introduces iCoTASC, a hybrid offline-online framework that breaks this trade-off. It leverages attribution-based importance to select per-dimension embeddings as a communication control signal, and models the diminishing semantic returns of quantization via a data-driven utility function. Crucially, utility lookup tables are precomputed offline for each transmitter, allowing real-time scheduling through simple table lookups and low-complexity refinements when channels vary.

iCoTASC supports distributed, multi-device systems without requiring retraining of the underlying task inference model. The framework's key innovation is shifting computational complexity offline while maintaining channel-adaptive performance online. For professionals building edge AI, IoT networks, or collaborative autonomous systems, this means semantic communication can now operate in real-time with minimal overhead, adapting to fluctuating wireless conditions. The approach promises significant reductions in latency and energy consumption for applications like drone swarms, smart factories, and real-time video analytics, where semantic understanding must survive bandwidth constraints.