Genal Activation: A Learnable Activation Function with Superior Generalization Across 16 Diverse Tasks

We introduce Genal Activation, a novel learnable activation function defined as Genal(x) = x · sigmoid(x/k), where k = softplus(θ) + ε and θ is a trainable parameter. Evaluated on 16 diverse benchmarks covering computer vision, medical diagnosis, physics-informed neural networks, NLP, reinforcement learning, and audio classification. Genal achieves 85.11% on CIFAR-10, 97.44% on Parkinson's diagnosis, and Navier-Stokes loss 44× lower than ReLU. Code: https://github.com/GenalFF/genal-activation