Describe how AdaGrad can be used to adaptively adjust the learning rate during training to overcome local minima and achieve faster convergence. Be sure to discuss any potential tradeoffs or limitations of this approach.

AdaGrad adapts the learning rate based on the entire history of squared gradients, which helps escape local minima by adjusting the step size for each parameter update. This can lead to faster convergence in non-convex spaces at the cost of potentially slower adaptation to changing gradient landscapes.