Implementación de una red neuronal generativa antagónica análoga para un sistema clasificador de patrones de colores rgb

Abstract

This paper documents the implementation of an analog generative adversarial system focused on creating new color combinations in the RGB color space. The system uses the outputs of an analog neural network previously trained to recognize the red, green, and blue channels, which act as the main inputs. From these signals, an internal dynamic is generated based on three functional blocks: voltage-controlled oscillators (VCOs), a white noise generator, and an analog mixer. These components interact competitively, perturbing and modulating the output signal in a non-deterministic manner, allowing the exploration of new configurations and emergent colors. The system's response was analyzed through simulations and parameter adjustments, demonstrating its ability to adapt and respond diversely to similar stimuli. This analog architecture represents a first step toward exploring physical generative networks, inspired by the variable behavior of biological systems