---

name: mas_regularized_training description: Apply the MAS quadratic regularizer during later-task training and log parameter drift evidence.

MAS Regularized Training

Use for later-task optimizer steps with task_loss + lambda * sum(Omega * (theta-theta_star)^2). Inputs are weights, task data, snapshot, importance, lambda, and optimizer settings. Outputs are updated parameters and loss/penalty traces. Validate with python tests/test_regularizer.py.