AI/ML+Physics Part 3: Designing an Architecture

Mass, momentum, and energy are conserved in our universe, leading to fundamental invariants that create symmetries in data. These symmetries, variances, and conservation laws are core principles in physics that can be incorporated into machine learning algorithms. For example, the laws of physics remain unchanged when translating or rotating objects.

In considering architecture choices, it is essential to enforce or promote these physical principles and discover new symmetries. This concept of physics is crucial not only for architecture design but also for defining loss functions and optimization algorithms used in training machine learning models.