We want to answer fundamental questions on the interface between Physics and Artificial Intelligence. Both fields share similar goals, for instance, finding structure in disordered data, but use different techniques to achieve these goals. In addition, the two research fields suffer from opposite shortcomings. For example, in Applied Physics, it is difficult to engineer optimal structures when there are too many degrees of freedom. In the field of Artificial Intelligence, and Machine Learning in particular, there is too little transparency into the machinery how certain results are obtained. Almost always, the physical intuition disappears in the end result. In this project, we want to combine the advantages of both fields, and solve important problems at their interface. On the one hand, we will develop a new kind of artificially intelligent algorithms that have internalized important physical concepts, and can therefore be used more efficiently in solving problems in Applied Physics. Such an artificial physicist could also be used to assist (neuro)-scientists and engineers in the development of theoretical models of physical and biological phenomena, and even to propose new experiments. On the other hand, we will use tools from Physics to gain a better understanding of the internal workings of artificial neural networks. A better physical understanding not only ensures more reliable proposed results, but will also lead to a more optimal training of the networks.
|Effective start/end date||1/11/20 → 31/10/22|
- Statistical Physics
- Artificial Neural Networks
Flemish discipline codes
- Machine learning and decision making
- Applied and interdisciplinary physics
- Artificial intelligence