In many sequential decision making applications, the change of decision would bring an additional cost, such as the wear-and-tear cost associated with changing server status. To control the switching cost, we introduce the problem of online convex optimization with continuous switching constraint, where the goal is to achieve a small regret given a budget on the \emph{overall} switching cost. We first investigate the hardness of the problem, and provide a lower bound of order $\Omega(\sqrt{T})$ when the switching cost budget $S=\Omega(\sqrt{T})$, and $\Omega(\min\{\frac{T}{S},T\})$ when $S=O(\sqrt{T})$, where $T$ is the time horizon. The essential idea is to carefully design an adaptive adversary, who can adjust the loss function according to the cumulative switching cost of the player incurred so far based on the orthogonal technique. We then develop a simple gradient-based algorithm which enjoys the minimax optimal regret bound. Finally, we show that, for strongly convex functions, the regret bound can be improved to $O(\log T)$ for $S=\Omega(\log T)$, and $O(\min\{T/\exp(S)+S,T\})$ for $S=O(\log T)$.
In this paper, we revisit the problem of smoothed online learning, in which the online learner suffers both a hitting cost and a switching cost, and target two performance metrics: competitive ratio and dynamic regret with switching cost. To bound the competitive ratio, we assume the hitting cost is known to the learner in each round, and investigate the greedy algorithm which simply minimizes the weighted sum of the hitting cost and the switching cost. Our theoretical analysis shows that the greedy algorithm, although straightforward, is $1+ \frac{2}{\alpha}$-competitive for $\alpha$-polyhedral functions, $1+O(\frac{1}{\lambda})$-competitive for $\lambda$-quadratic growth functions, and $1 + \frac{2}{\sqrt{\lambda}}$-competitive for convex and $\lambda$-quadratic growth functions. To bound the dynamic regret with switching cost, we follow the standard setting of online convex optimization, in which the hitting cost is convex but hidden from the learner before making predictions. We modify Ader, an existing algorithm designed for dynamic regret, slightly to take into account the switching cost when measuring the performance. The proposed algorithm, named as Smoothed Ader, attains an optimal $O(\sqrt{T(1+P_T)})$ bound for dynamic regret with switching cost, where $P_T$ is the path-length of the comparator sequence. Furthermore, if the hitting cost is accessible in the beginning of each round, we obtain a similar guarantee without the bounded gradient condition.
\underline{D}eep \underline{A}UC (area under the ROC curve) \underline{M}aximization (DAM) has attracted much attention recently due to its great potential for imbalanced data classification. However, the research on \underline{F}ederated \underline{D}eep \underline{A}UC \underline{M}aximization (FDAM) is still limited. Compared with standard federated learning (FL) approaches that focus on decomposable minimization objectives, FDAM is more complicated due to its minimization objective is non-decomposable over individual examples. In this paper, we propose improved FDAM algorithms for heterogeneous data by solving the popular non-convex strongly-concave min-max formulation of DAM in a distributed fashion. A striking result of this paper is that the communication complexity of the proposed algorithm is a constant independent of the number of machines and also independent of the accuracy level, which improves an existing result by orders of magnitude. Of independent interest, the proposed algorithm can also be applied to a class of non-convex-strongly-concave min-max problems. The experiments have demonstrated the effectiveness of our FDAM algorithm on benchmark datasets, and on medical chest X-ray images from different organizations. Our experiment shows that the performance of FDAM using data from multiple hospitals can improve the AUC score on testing data from a single hospital for detecting life-threatening diseases based on chest radiographs.
In this paper~\footnote{The original title is "Momentum SGD with Robust Weighting For Imbalanced Classification"}, we present a simple yet effective method (ABSGD) for addressing the data imbalance issue in deep learning. Our method is a simple modification to momentum SGD where we leverage an attentional mechanism to assign an individual importance weight to each gradient in the mini-batch. Unlike existing individual weighting methods that learn the individual weights by meta-learning on a separate balanced validation data, our weighting scheme is self-adaptive and is grounded in distributionally robust optimization. The weight of a sampled data is systematically proportional to exponential of a scaled loss value of the data, where the scaling factor is interpreted as the regularization parameter in the framework of information-regularized distributionally robust optimization. We employ a step damping strategy for the scaling factor to balance between the learning of feature extraction layers and the learning of the classifier layer. Compared with exiting meta-learning methods that require three backward propagations for computing mini-batch stochastic gradients at three different points at each iteration, our method is more efficient with only one backward propagation at each iteration as in standard deep learning methods. Compared with existing class-level weighting schemes, our method can be applied to online learning without any knowledge of class prior, while enjoying further performance boost in offline learning combined with existing class-level weighting schemes. Our empirical studies on several benchmark datasets also demonstrate the effectiveness of our proposed method
Deep AUC Maximization (DAM) is a paradigm for learning a deep neural network by maximizing the AUC score of the model on a dataset. Most previous works of AUC maximization focus on the perspective of optimization by designing efficient stochastic algorithms, and studies on generalization performance of DAM on difficult tasks are missing. In this work, we aim to make DAM more practical for interesting real-world applications (e.g., medical image classification). First, we propose a new margin-based surrogate loss function for the AUC score (named as the AUC margin loss). It is more robust than the commonly used AUC square loss, while enjoying the same advantage in terms of large-scale stochastic optimization. Second, we conduct empirical studies of our DAM method on difficult medical image classification tasks, namely classification of chest x-ray images for identifying many threatening diseases and classification of images of skin lesions for identifying melanoma. Our DAM method has achieved great success on these difficult tasks, i.e., the 1st place on Stanford CheXpert competition (by the paper submission date) and Top 1% rank (rank 33 out of 3314 teams) on Kaggle 2020 Melanoma classification competition. We also conduct extensive ablation studies to demonstrate the advantages of the new AUC margin loss over the AUC square loss on benchmark datasets. To the best of our knowledge, this is the first work that makes DAM succeed on large-scale medical image datasets.
Adam is a widely used stochastic optimization method for deep learning applications. While practitioners prefer Adam because it requires less parameter tuning, its use is problematic from a theoretical point of view since it may not converge. Variants of Adam have been proposed with provable convergence guarantee, but they tend not be competitive with Adam on the practical performance. In this paper, we propose a new method named Adam$^+$ (pronounced as Adam-plus). Adam$^+$ retains some of the key components of Adam but it also has several noticeable differences: (i) it does not maintain the moving average of second moment estimate but instead computes the moving average of first moment estimate at extrapolated points; (ii) its adaptive step size is formed not by dividing the square root of second moment estimate but instead by dividing the root of the norm of first moment estimate. As a result, Adam$^+$ requires few parameter tuning, as Adam, but it enjoys a provable convergence guarantee. Our analysis further shows that Adam$^+$ enjoys adaptive variance reduction, i.e., the variance of the stochastic gradient estimator reduces as the algorithm converges, hence enjoying an adaptive convergence. We also propose a more general variant of Adam$^+$ with different adaptive step sizes and establish their fast convergence rate. Our empirical studies on various deep learning tasks, including image classification, language modeling, and automatic speech recognition, demonstrate that Adam$^+$ significantly outperforms Adam and achieves comparable performance with best-tuned SGD and momentum SGD.
Off-policy policy optimization is a challenging problem in reinforcement learning (RL). The algorithms designed for this problem often suffer from high variance in their estimators, which results in poor sample efficiency, and have issues with convergence. A few variance-reduced on-policy policy gradient algorithms have been recently proposed that use methods from stochastic optimization to reduce the variance of the gradient estimate in the REINFORCE algorithm. However, these algorithms are not designed for the off-policy setting and are memory-inefficient, since they need to collect and store a large ``reference'' batch of samples from time to time. To achieve variance-reduced off-policy-stable policy optimization, we propose an algorithm family that is memory-efficient, stochastically variance-reduced, and capable of learning from off-policy samples. Empirical studies validate the effectiveness of the proposed approaches.
This paper focuses on stochastic methods for solving smooth non-convex strongly-concave min-max problems, which have received increasing attention due to their potential applications in deep learning (e.g., deep AUC maximization). However, most of the existing algorithms are slow in practice, and their analysis revolves around the convergence to a nearly stationary point. We consider leveraging the Polyak-\L ojasiewicz (PL) condition to design faster stochastic algorithms with stronger convergence guarantee. Although PL condition has been utilized for designing many stochastic minimization algorithms, their applications for non-convex min-max optimization remains rare. In this paper, we propose and analyze proximal epoch-based methods, and establish fast convergence in terms of both {\bf the primal objective gap and the duality gap}. Our analysis is interesting in threefold: (i) it is based on a novel Lyapunov function that consists of the primal objective gap and the duality gap of a regularized function; (ii) it only requires a weaker PL condition for establishing the primal objective convergence than that required for the duality gap convergence; (iii) it yields the optimal dependence on the accuracy level $\epsilon$, i.e., $O(1/\epsilon)$. We also make explicit the dependence on the problem parameters and explore regions of weak convexity parameter that lead to improved dependence on condition numbers. Experiments on deep AUC maximization demonstrate the effectiveness of our methods. Our method also beats the 1st place on {\bf Stanford CheXpert competition} in terms of AUC on the public validation set.
In this paper, we propose a practical online method for solving a distributionally robust optimization (DRO) for deep learning, which has important applications in machine learning for improving the robustness of neural networks. In the literature, most methods for solving DRO are based on stochastic primal-dual methods. However, primal-dual methods for deep DRO suffer from several drawbacks: (1) manipulating a high-dimensional dual variable corresponding to the size of data is time expensive; (2) they are not friendly to online learning where data is coming sequentially. To address these issues, we transform the min-max formulation into a minimization formulation and propose a practical duality-free online stochastic method for solving deep DRO with KL divergence regularization. The proposed online stochastic method resembles the practical stochastic Nesterov's method in several perspectives that are widely used for learning deep neural networks. Under a Polyak-Lojasiewicz (PL) condition, we prove that the proposed method can enjoy an optimal sample complexity and a better round complexity (the number of gradient evaluations divided by a fixed mini-batch size) with a moderate mini-batch size than existing algorithms for solving the min-max or min formulation of DRO. Of independent interest, the proposed method can be also used for solving a family of stochastic compositional problems.
In this paper, we propose robust stochastic algorithms for solving convex compositional problems of the form $f(\E_\xi g(\cdot; \xi)) + r(\cdot)$ by establishing {\bf sub-Gaussian confidence bounds} under weak assumptions about the tails of noise distribution, i.e., {\bf heavy-tailed noise} with bounded second-order moments. One can achieve this goal by using an existing boosting strategy that boosts a low probability convergence result into a high probability result. However, piecing together existing results for solving compositional problems suffers from several drawbacks: (i) the boosting technique requires strong convexity of the objective; (ii) it requires a separate algorithm to handle non-smooth $r$; (iii) it also suffers from an additional polylogarithmic factor of the condition number. To address these issues, we directly develop a single-trial stochastic algorithm for minimizing optimal strongly convex compositional objectives, which has a nearly optimal high probability convergence result matching the lower bound of stochastic strongly convex optimization up to a logarithmic factor. To the best of our knowledge, this is the first work that establishes nearly optimal sub-Gaussian confidence bounds for compositional problems under heavy-tailed assumptions.