Escaping Saddle-Point Faster under Interpolation-like Conditions