.png){kind=icon}
.png){kind=icon}
Return
Enhancing randomized recurrent neural networks with explainable attribution methods
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.132318.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• We propose a method to enhance randomized recurrent neural networks by weighting their hidden states using attribution scores derived from explainable AI techniques. • Theoretical analysis shows that our proposed aggregation corresponds to a second-order residual RNN formulation, whose Jacobian spectral radius is automatically adapted based on XAI-driven attribution values. • Aggregating hidden states using XAI-derived attribution scores improves classification performance across several datasets, outperforming standard baselines, especially under noisy conditions.