A new artificial intelligence (AI) technology that could discover hidden physical laws has been developed by researchers at Kobe University and Osaka University. The AI can extract hidden equations of motion from regular observational data, which is then used to create a model based on the laws of physics.
The new development could enable experts to discover the hidden equations of motion behind phenomena that are unexplainable.
The research team included Associate Professor Yaguchi Takaharu and Ph.D. student Chen Yuhan from Kobe University, as well as Associate Professor Matsubara Takashi from Osaka University.
The research was presented last month at the Thirty-fifth Conference on Neural Information Processing Systems (NeurlPS2021).
Predicting Physical Phenomena
To make predictions on physical phenomena, experts usually rely on simulations with supercomputers. The simulations use mathematical models based on the laws of physics, but the results can be unreliable if the model is questionable. This is why it is crucial to have a method of producing reliable models from the observational data of phenomena.
The new research developed a method of discovering novel equations of motion in observational data. Previous research has focused on discovering equations of motion from data, but some required the data to be in the appropriate format. The problem is that there are many instances where experts don’t know the best data format to use, so it is difficult to apply realistic data.
Illuminating Unknown Geometric Properties
The researchers addressed this challenge by illuminating the unknown geometric properties behind phenomena. This enabled them to develop an AI that can find these geometric properties in data. If the AI can extract equations of motion from data, then the equations could be used to create models and simulations that follow physical laws.
Physical simulations take place in fields like weather forecasting, drug discovery, and car design. However, they usually require extensive calculation. If AI can learn from the data of specific phenomena, as well as construct small-scale models using the new method, then the calculations could be simplified, sped up, and faithful to the laws of physics.
The method could also be applied to areas unrelated to physics, enabling physics knowledge-based investigations and simulations for phenomena previously considered impossible to explain. One such example is that it could be used to find a hidden equation of motion in animal population data that shows the change in the number of individuals, which could help provide insights into ecosystem sustainability.