During batch training of a neural network, you observe oscillating loss values. What modifications should you make to your model to achieve convergence?