Adaptive optimization methods, which perform local optimization with a metric constructed from the history of iterates, are becoming increasingly popular for training deep neural networks. Examples include AdaGrad, RMSProp, and Adam. We show that for simple overparameterized problems, adaptive methods often find drastically different solutions than gradient descent (GD) or stochastic gradient descent (SGD). We construct an illustrative binary classification problem where the data is linearly separable, GD and SGD achieve zero test error, and AdaGrad, Adam, and RMSProp attain test errors arbitrarily close to half. We additionally study the empirical generalization capability of adaptive methods on several state-of-the-art deep learning models. We observe that the solutions found by adaptive methods generalize worse (often significantly worse) than SGD, even when these solutions have better training performance. These results suggest that practitioners should reconsider the use of adaptive methods to train neural networks.

Adaptive stochastic gradient methods such as AdaGrad have gained popularity in particular for training deep neural networks. The most commonly used and studied variant maintains a diagonal matrix approximation to second order information by accumulating past gradients which are used to tune the step size adaptively. In certain situations the full-matrix variant of AdaGrad is expected to attain better performance, however in high dimensions it is computationally impractical. We present Ada-LR and RadaGrad two computationally efficient approximations to full-matrix AdaGrad based on randomized dimensionality reduction. They are able to capture dependencies between features and achieve similar performance to full-matrix AdaGrad but at a much smaller computational cost. We show that the regret of Ada-LR is close to the regret of full-matrix AdaGrad which can have an up-to exponentially smaller dependence on the dimension than the diagonal variant. Empirically, we show that Ada-LR and RadaGrad perform similarly to full-matrix AdaGrad. On the task of training convolutional neural networks as well as recurrent neural networks, RadaGrad achieves faster convergence than diagonal AdaGrad.

Thus, diagonal preconditioning methods remain popular.

Conversely, adaptive first order methods are very popular in Machine Learning, with AdaGrad, [12],beingthemostprominent methodamongthisclass. AdaGrad isanonlinelearning algorithm which adapts its learning rate using the feedback (gradients) received through the optimization process, and is known to successfully handle noisy feedback.

Con set isa (scaled)`p ball, 1 p< 2, then, considering(2), the thebestmethod p d/logdtimeslar weprovidene . Formally, if 2 then s 2 { 1}d implies (sj j)j d 2 .

Figure 2: Testlog-perplexityofa Transformer -Bigmodelon WMT'14 en!fr, whentrainingwith batchsizesof 384 (left) and 768 (right).

We study the implicit bias of AdaGrad on separable linear classification problems.