once and zeros

The installation will act as an abstract data visualisation of the learning process of a neural network transferred back to the training materials as a displacement in space and light.

The underlying modified algorithm will trace back the impact of the modifications of the "weights and biases" in the network as input to the neural network in its learning phase and displays the input pixels in kinetic motion and tinting its shape of light.

Each Cube represents a pixel of the input data of the network.

This process is normally hidden in neural networks and stays unlogged. So an attribution of a given result of a trained network cannot be traced back to the input data. This results in users having to blindly believe in the given answer of a result, kept away from the possibility of understanding its true origin.