Second-Order Optimization for Non-Convex Machine Learning: An Empirical Study
This addresses optimization challenges for researchers and practitioners in non-convex ML, but it is incremental as it focuses on empirical evaluation of existing second-order methods.
The paper tackles the deficiencies of first-order optimization methods like SGD in non-convex machine learning by empirically evaluating Newton-type methods, showing they are computationally competitive with hand-tuned SGD, achieve comparable or better generalization, and are robust to hyper-parameter settings while escaping flat regions and saddle points.
While first-order optimization methods such as stochastic gradient descent (SGD) are popular in machine learning (ML), they come with well-known deficiencies, including relatively-slow convergence, sensitivity to the settings of hyper-parameters such as learning rate, stagnation at high training errors, and difficulty in escaping flat regions and saddle points. These issues are particularly acute in highly non-convex settings such as those arising in neural networks. Motivated by this, there has been recent interest in second-order methods that aim to alleviate these shortcomings by capturing curvature information. In this paper, we report detailed empirical evaluations of a class of Newton-type methods, namely sub-sampled variants of trust region (TR) and adaptive regularization with cubics (ARC) algorithms, for non-convex ML problems. In doing so, we demonstrate that these methods not only can be computationally competitive with hand-tuned SGD with momentum, obtaining comparable or better generalization performance, but also they are highly robust to hyper-parameter settings. Further, in contrast to SGD with momentum, we show that the manner in which these Newton-type methods employ curvature information allows them to seamlessly escape flat regions and saddle points.