Ordinary neural networks learn from examples, like a person memorizing tunes by ear. But networks with physical laws embedded from the start resemble a musician who knows notes and harmony: they can play a new piece from just a few chords. Instead of mindless memorization, they use equations like energy conservation or motion. So their solutions always obey reality, even with scarce data.
This approach is indispensable when studying distant galaxies or dark matter: the network accurately predicts the system’s behavior after seeing just 10% of the necessary data. Physics guides it like a score guides a musician. This accelerates discoveries, from gravitational waves to digital twins of complex systems. The ideas of Newton, von Neumann, and Emmy Noether merge here: physics provides the rules, computation embodies them, and symmetries tie everything together.
Accurate predictions are possible even after training on just 10% of the data — physics won’t let it go wrong.
It’s reminiscent of ‘The Matrix’: knowing the laws of reality gives you power, but here we’re not breaking the world, we’re learning about it.