2024/12/13 by Júnior, José Gilberto Barbosa de Medeiros, André Guarnier de Mitri, de Mitri, Andre Guarnier +2
Computer Science · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Neural Networks and Applications #Time Series Analysis and Forecasting
paper · pdf · doi:10.48550/arxiv.2412.09889
openalex publication_date 2024/12/13 · openalex created_date 2025/10/24 · openalex updated_date 2026/07/28
This paper investigates the lack of research on activation functions for\nneural network models in time series tasks. It highlights the need to identify\nessential properties of these activations to improve their effectiveness in\nspecific domains. To this end, the study comprehensively analyzes properties,\nsuch as bounded, monotonic, nonlinearity, and periodicity, for activation in\ntime series neural networks. We propose a new activation that maximizes the\ncoverage of these properties, called LeakySineLU. We empirically evaluate the\nLeakySineLU against commonly used activations in the literature using 112\nbenchmark datasets for time series classification, obtaining the best average\nranking in all comparative scenarios.\n