Learning Unstable

Parameter tuning to stabilize the learning