Beyond Lipschitz: Sharp Generalization and Excess Risk Bounds for Full-Batch GD
This addresses theoretical limitations in optimization theory for machine learning practitioners, offering improved bounds for full-batch GD with broad implications for nonconvex, convex, and strongly convex settings.
The paper provides sharp generalization and excess risk bounds for full-batch Gradient Descent on smooth losses, bypassing Lipschitz assumptions and showing that for convex losses, generalization error is tighter than SGD bounds, and for strongly convex losses, it achieves similar excess risk rates with exponentially fewer iterations.
We provide sharp path-dependent generalization and excess risk guarantees for the full-batch Gradient Descent (GD) algorithm on smooth losses (possibly non-Lipschitz, possibly nonconvex). At the heart of our analysis is an upper bound on the generalization error, which implies that average output stability and a bounded expected optimization error at termination lead to generalization. This result shows that a small generalization error occurs along the optimization path, and allows us to bypass Lipschitz or sub-Gaussian assumptions on the loss prevalent in previous works. For nonconvex, convex, and strongly convex losses, we show the explicit dependence of the generalization error in terms of the accumulated path-dependent optimization error, terminal optimization error, number of samples, and number of iterations. For nonconvex smooth losses, we prove that full-batch GD efficiently generalizes close to any stationary point at termination, and recovers the generalization error guarantees of stochastic algorithms with fewer assumptions. For smooth convex losses, we show that the generalization error is tighter than existing bounds for SGD (up to one order of error magnitude). Consequently the excess risk matches that of SGD for quadratically less iterations. Lastly, for strongly convex smooth losses, we show that full-batch GD achieves essentially the same excess risk rate as compared with the state of the art on SGD, but with an exponentially smaller number of iterations (logarithmic in the dataset size).