Various Invariance Based Algorithm

Invariance-based algorithms aim to improve machine learning models' robustness and generalization by identifying and exploiting transformations that leave the underlying data semantics unchanged. Current research focuses on developing methods that learn these invariances from data, either implicitly through architecture design (e.g., using pruning to discover invariant subnetworks) or explicitly by incorporating invariance constraints into the learning process (e.g., enforcing Galilean or Lorentz invariance in PDE discovery). This approach is proving valuable across diverse applications, including robotics, natural language processing (through efficient quantization), and scientific discovery (e.g., identifying governing equations from noisy data), leading to more efficient and generalizable models.