Learning Physics
We use machine learning to bring physical model buidling a step further. We believe that by formulating the scientific method in a way that is amenable to an automated learning process, we will be able to tackle problems that are intractable today.
Our vision
The general workflow of scientific research can be cast into a simple sequence of steps. First, we make an observation. We then formulate a set of rules which describe this observation. Finally, we try to see if our theory is of general validity by testing our rules in situations slightly different from the initial observation.
Newton’s anecdotal observation of a falling apple made him formulate his equations of motion. Using his rules, we can now calculate the trajectory of a spaceship flying to the Mars, a situation very different from a falling apple.
Over the last few hundred years, physicist developed a powerful toolbox for the formulation of such theories. Despite the apparent complexity of the challenges we deal with, ranging from the formation of galactic structures in our universe down to the microscopic building blocks of matter, all our theories describe somewhat simple systems. After all, our theories all fit onto one or two lines of manageable formulas. However, we believe that in order to fill in the remaining voids of our understanding, we need to deal with complexities that go beyond what a human mind can embrace and cast onto the famous “back of the envelope”.
To take the necessary steps forward we explore the use of artificial intelligence in physical model building. Our goal is to automate the very act of formulating theories. We tackle this goal step-by-step. We first try to “replace” the giants on whose shoulders we stand by trying to rediscover the basic theories underlying our current understanding of physics using learning machines. The next steps in our endeavour will be to apply our artificial-intelligence based framework to the description of highly complex systems like frustrated quantum magnets, quantum systems at the quantum-to-classical threshold or strongly interacting quantum systems far from equilibrium.