
A novel adaptive activation function with an even cubic nonlinearity is introduced that enhances the accuracy of neural networks without substantial additional computational resources, while exhibiting a tradeoff between convergence and accuracy.


coremsg

enhancing-neural-network-accuracy-with-a-novel-adaptive-activation-function


Enhancing Neural Network Accuracy with a Novel Adaptive Activation Function


title_rewrite


This paper proposes a novel feature-based echo-state network (Feat-ESN) architecture that uses smaller parallel reservoirs driven by different input feature combinations to significantly reduce the computational complexity of traditional ESNs while maintaining predictive performance.


feature-based-echo-state-networks-an-interpretable-and-efficient-approach-to-reservoir-computing-for-time-series-prediction


Feature-Based Echo-State Networks: An Interpretable and Efficient Approach to Reservoir Computing for Time-Series Prediction
