v0.1.1

EvoActiv, an open-source framework for the evolutionary discovery of neural network activation
functions. In contrast to functions designed manually or via gradient-based search, EvoActiv navigates the
space of possible activations using symbolic genetic evolution. The framework operates by generating,
evaluating, and selecting functions based on their performance on a given task. We conduct experiments on
the MNIST and Fashion-MNIST datasets, demonstrating that EvoActiv can autonomously discover novel
activation functions that outperform standard baselines like ReLU and Swish in terms of final accuracy and
convergence speed. Our results indicate that evolutionary discovery is a viable methodology for generating
novel, mathematically interpretable activation functions that can enhance model performance.

Keywords: Neural networks, activation functions, genetic programming, evolutionary algorithms, sym
bolic regression