Generalization and optimization guarantees on the population loss often rely on uniform convergence based analysis, typically based on the Rademacher complexity of the predictors. The rich representation power of modern models has led to concerns about this approach. In this paper, we present generalization and optimization guarantees in terms of the complexity of the gradients, as measured by the Loss Gradient Gaussian Width (LGGW). First, we introduce generalization guarantees directly in terms of the LGGW under a flexible gradient domination condition. Second, we show that sample reuse in ERM does not make the empirical gradients deviate from the population gradients as long as the LGGW is small. Third, focusing on deep networks, we bound their single-sample LGGW in terms of the Gaussian width of the featurizer, i.e., the output of the last-but-one layer. To our knowledge, our generalization and optimization guarantees in terms of LGGW are the first results of its kind, and hold considerable promise towards quantitatively tight bounds for deep models.

Recent years have seen advances in generalization bounds for noisy stochastic algorithms, especially stochastic gradient Langevin dynamics (SGLD) based on stability (Mou et al., 2018; Li et al., 2020) and information theoretic approaches (Xu & Raginsky, 2017; Negrea et al., 2019; Steinke & Zakynthinou, 2020). In this paper, we unify and substantially generalize stability based generalization bounds and make three technical contributions. First, we bound the generalization error in terms of expected (not uniform) stability which arguably leads to quantitatively sharper bounds. Second, as our main contribution, we introduce Exponential Family Langevin Dynamics (EFLD), a substantial generalization of SGLD, which includes noisy versions of Sign-SGD and quantized SGD as special cases. We establish data dependent expected stability based generalization bounds for any EFLD algorithm with a O(1/n) sample dependence and dependence on gradient discrepancy rather than the norm of gradients, yielding significantly sharper bounds. Third, we establish optimization guarantees for special cases of EFLD. Further, empirical results on benchmarks illustrate that our bounds are non-vacuous, quantitatively sharper than existing bounds, and behave correctly under noisy labels.

Recent empirical work on stochastic gradient descent (SGD) applied to over-parameterized deep learning has shown that most gradient components over epochs are quite small. Inspired by such observations, we rigorously study properties of Truncated SGD (T-SGD), that truncates the majority of small gradient components to zeros. Considering non-convex optimization problems, we show that the convergence rate of T-SGD matches the order of vanilla SGD. We also establish the generalization error bound for T-SGD. Further, we propose Noisy Truncated SGD (NT-SGD), which adds Gaussian noise to the truncated gradients. We prove that NT-SGD has the same convergence rate as T-SGD for non-convex optimization problems. We demonstrate that with the help of noise, NT-SGD can provably escape from saddle points and requires less noise compared to previous related work. We also prove that NT-SGD achieves better generalization error bound compared to T-SGD because of the noise. Our generalization analysis is based on uniform stability and we show that additional noise in the gradient update can boost the stability. Our experiments on a variety of benchmark datasets with various network architectures verify that NT-SGD matches the speed and accuracy of vanilla SGD while effectively working with sparse gradients, and can successfully escape poor local minima.

Adaptive gradient methods such as AdaGrad, RMSprop and Adam have been optimizers of choice for deep learning due to their fast training speed. However, it was recently observed that their generalization performance is often worse than that of SGD for over-parameterized neural networks. While new algorithms such as AdaBound, SWAT, and Padam were proposed to improve the situation, the provided analyses are only committed to optimization bounds for the training objective, leaving critical generalization capacity unexplored. To close this gap, we propose \textit{\textbf{S}table \textbf{A}daptive \textbf{G}radient \textbf{D}escent} (\textsc{SAGD}) for nonconvex optimization which leverages differential privacy to boost the generalization performance of adaptive gradient methods. Theoretical analyses show that \textsc{SAGD} has high-probability convergence to a population stationary point. We further conduct experiments on various popular deep learning tasks and models. Experimental results illustrate that \textsc{SAGD} is empirically competitive and often better than baselines.

While stochastic gradient descent (SGD) and variants have been surprisingly successful for training deep nets, several aspects of the optimization dynamics and generalization are still not well understood. In this paper, we present new empirical observations and theoretical results on both the optimization dynamics and generalization behavior of SGD for deep nets based on the Hessian of the training loss and associated quantities. We consider three specific research questions: (1) what is the relationship between the Hessian of the loss and the second moment of stochastic gradients (SGs)? (2) how can we characterize the stochastic optimization dynamics of SGD with fixed step sizes based on the first and second moments of SGs? and (3) how can we characterize a scale-invariant generalization bound of deep nets based on the Hessian of the loss? Throughout the paper, we support theoretical results with empirical observations, with experiments on synthetic data, MNIST, and CIFAR-10, with different batch sizes, and with different difficulty levels by synthetically adding random labels.

With the rapid growth of Internet technologies, cloud computing and social networks have become ubiquitous. An increasing number of people participate in social networks and massive online social data are obtained. In order to exploit knowledge from copious amounts of data obtained and predict social behavior of users, we urge to realize data mining in social networks. Almost all online websites use cloud services to effectively process the large scale of social data, which are gathered from distributed data centers. These data are so large-scale, high-dimension and widely distributed that we propose a distributed sparse online algorithm to handle them. Additionally, privacy-protection is an important point in social networks. We should not compromise the privacy of individuals in networks, while these social data are being learned for data mining. Thus we also consider the privacy problem in this article. Our simulations shows that the appropriate sparsity of data would enhance the performance of our algorithm and the privacy-preserving method does not significantly hurt the performance of the proposed algorithm.

With the rapid growth in multimedia services and the enormous offers of video content in online social networks, users have difficulty in obtaining their interests. Therefore, various personalized recommendation systems have been proposed. However, they ignore that the accelerated proliferation of social media data has led to the big data era, which has greatly impeded the process of video recommendation. In addition, none of them has considered both the privacy of users' contexts (e.g., social status, ages, and hobbies) and video service vendors' repositories, which are extremely sensitive and of significant commercial value. To handle these problems, we propose a cloud-assisted differentially private video recommendation system based on distributed online learning. In our framework, service vendors are modeled as distributed cooperative learners, recommending videos according to user's context, while simultaneously adapting the video-selection strategy based on user-click feedback to maximize total user clicks (reward). Considering the sparsity and heterogeneity of big social media data, we also propose a novel geometric differentially private model, which can greatly reduce the performance loss. Our simulation shows the proposed algorithms outperform other existing methods and keep a delicate balance between the total reward and privacy preserving level.