On Computationally Efficient Learning For Stabilizers And Beyond