Deepspace is a deep convolutional Neural Network designed to create deep space imagery based on Hubble’s resource gallery. We trained the network with stunning imagery created by the Hubble Space Telescope, which was launced into space aboard spaceshuttle Discovery on April 24th, 1990.

samples from dataset

In fusing this visual data with the power of machine leraning we created a tool which extracted a “formula” on how to visually recreate a galaxy. Due to our personal connection to this field and it’s exploration, this project is a very meaningful one. It allows us to take one step closer towards the big unknown, to experience realities which may be out there somewhere but are hidden from our human experience. It makes us wonder how far a combination of human effort and machine learning might take us one day.

process-video of network training

We use these tools to create art today but very similar tools will be used one day to help us design a craft that can transport us to far away places, or a tool to bring far away places closer to us.