X-RACE Framework Prunes LSTM Units via One-Shot Dual Optimization
September 11, 2026
X-RACE utilizes a low-complexity, one-shot dual-optimization strategy to simultaneously prune irrelevant input subcarriers and internal hidden units in LSTMs. It introduces temporal XAI metrics like Saturation Time and Importance Drift to characterize memory convergence in channel estimation tasks.
HOW THIS AFFECTS YOU
●
researcherYou can use these new temporal metrics to better characterize LSTM learning dynamics.