Artificial Neural Networks are one of the most used subsets of Machine Learning systems. They are inspired by the neural structure of the biological brain and consist of a series of layered and interconnected Artificial Neurons — Perceptrons. As Artificial Neural Networks grow deeper, their number of layers increase, as can happen in the number of Perceptrons per layer. This increases the number of decision iterations, developing an iterative complexity in the decision-making process and turning these systems into black boxes. The illustrations of the inner workings of these systems, coming from the scientific community, also lack explainability and connection with the non-specialist observer.

TransparentPerceptron is an algorithmic data-driven visualization that reveals the accumulated decision iterations of the most elementary system of an Artificial Neural Network: a Perceptron. The single data point and the goal to be achieved are both generated randomly at the startup of this simplified system. Each iteration is less blurred as it approaches the system’s goal.