A common strategy to train deep neural networks (DNNs) is to use very large architectures and to train them until they (almost) achieve zero training error. Empirically observed good generalization performance on test data, even in the presence of lots of label noise, corroborate such a procedure. On the other hand, in statistical learning theory it is known that over-fitting models may lead to poor generalization properties, occurring in e.g. empirical risk minimization (ERM) over too large hypotheses classes. Inspired by this contradictory behavior, so-called interpolation methods have recently received much attention, leading to consistent and optimally learning methods for some local averaging schemes with zero training error. However, there is no theoretical analysis of interpolating ERM-like methods so far. We take a step in this direction by showing that for certain, large hypotheses classes, some interpolating ERMs enjoy very good statistical guarantees while others fail in the worst sense. Moreover, we show that the same phenomenon occurs for DNNs with zero training error and sufficiently large architectures.

Online learning algorithms update models via one sample per iteration, thus efficient to process large-scale datasets and useful to detect malicious events for social benefits, such as disease outbreak and traffic congestion on the fly. However, existing algorithms for graph-structured models focused on the offline setting and the least square loss, incapable for online setting, while methods designed for online setting cannot be directly applied to the problem of complex (usually non-convex) graph-structured sparsity model. To address these limitations, in this paper we propose a new algorithm for graph-structured sparsity constraint problems under online setting, which we call \textsc{GraphDA}. The key part in \textsc{GraphDA} is to project both averaging gradient (in dual space) and primal variables (in primal space) onto lower dimensional subspaces, thus capturing the graph-structured sparsity effectively. Furthermore, the objective functions assumed here are generally convex so as to handle different losses for online learning settings. To the best of our knowledge, \textsc{GraphDA} is the first online learning algorithm for graph-structure constrained optimization problems. To validate our method, we conduct extensive experiments on both benchmark graph and real-world graph datasets. Our experiment results show that, compared to other baseline methods, \textsc{GraphDA} not only improves classification performance, but also successfully captures graph-structured features more effectively, hence stronger interpretability.

Short-term demand forecasting models commonly combine convolutional and recurrent layers to extract complex spatiotemporal patterns in data. Long-term histories are also used to consider periodicity and seasonality patterns as time series data. In this study, we propose an efficient architecture, Temporal-Guided Network (TGNet), which utilizes graph networks and temporal-guided embedding. Graph networks extract invariant features to permutations of adjacent regions instead of convolutional layers. Temporal-guided embedding explicitly learns temporal contexts from training data and is substituted for the input of long-term histories from days/weeks ago. TGNet learns an autoregressive model, conditioned on temporal contexts of forecasting targets from temporal-guided embedding. Finally, our model achieves competitive performances with other baselines on three spatiotemporal demand dataset from real-world, but the number of trainable parameters is about 20 times smaller than a state-of-the-art baseline. We also show that temporal-guided embedding learns temporal contexts as intended and TGNet has robust forecasting performances even to atypical event situations.

Multi-view clustering has received much attention recently. Most of the existing multi-view clustering methods only focus on one-sided clustering. As the co-occurring data elements involve the counts of sample-feature co-occurrences, it is more efficient to conduct two-sided clustering along the samples and features simultaneously. To take advantage of two-sided clustering for the co-occurrences in the scene of multi-view clustering, a two-sided multi-view clustering method is proposed, i.e., multi-view information-theoretic co-clustering (MV-ITCC). The proposed method realizes two-sided clustering for co-occurring multi-view data under the formulation of information theory. More specifically, it exploits the agreement and disagreement among views by sharing a common clustering results along the sample dimension and keeping the clustering results of each view specific along the feature dimension. In addition, the mechanism of maximum entropy is also adopted to control the importance of different views, which can give a right balance in leveraging the agreement and disagreement. Extensive experiments are conducted on text and image multi-view datasets. The results clearly demonstrate the superiority of the proposed method.

We study the problem of fitting an ultrametric distance to a dissimilarity graph in the context of hierarchical cluster analysis. Standard hierarchical clustering methods are specified procedurally, rather than in terms of the cost function to be optimized. We aim to overcome this limitation by presenting a general optimization framework for ultrametric fitting. Our approach consists of modeling the latter as a constrained optimization problem over the continuous space of ultrametrics. So doing, we can leverage the simple, yet effective, idea of replacing the ultrametric constraint with an equivalent min-max operation injected directly into the cost function. The proposed reformulation leads to an unconstrained optimization problem that can be efficiently solved by gradient descent methods. The flexibility of our framework allows us to investigate several cost functions, following the classic paradigm of combining a data fidelity term with a regularization. While we provide no theoretical guarantee to find the global optimum, the numerical results obtained over a number of synthetic and real datasets demonstrate the good performance of our approach with respect to state-of-the-art agglomerative algorithms. This makes us believe that the proposed framework sheds new light on the way to design a new generation of hierarchical clustering methods.

In this work we propose Hebbian-descent as a biologically plausible learning rule for hetero-associative as well as auto-associative learning in single layer artificial neural networks. It can be used as a replacement for gradient descent as well as Hebbian learning, in particular in online learning, as it inherits their advantages while not suffering from their disadvantages. We discuss the drawbacks of Hebbian learning as having problems with correlated input data and not profiting from seeing training patterns several times. For gradient descent we identify the derivative of the activation function as problematic especially in online learning. Hebbian-descent addresses these problems by getting rid of the activation function's derivative and by centering, i.e. keeping the neural activities mean free, leading to a biologically plausible update rule that is provably convergent, does not suffer from the vanishing error term problem, can deal with correlated data, profits from seeing patterns several times, and enables successful online learning when centering is used. We discuss its relationship to Hebbian learning, contrastive learning, and gradient decent and show that in case of a strictly positive derivative of the activation function Hebbian-descent leads to the same update rule as gradient descent but for a different loss function. In this case Hebbian-descent inherits the convergence properties of gradient descent, but we also show empirically that it converges when the derivative of the activation function is only non-negative, such as for the step function for example. Furthermore, in case of the mean squared error loss Hebbian-descent can be understood as the difference between two Hebb-learning steps, which in case of an invertible and integrable activation function actually optimizes a generalized linear model. ...

We revisit the notion of individual fairness first proposed by Dwork et al. [2012], which asks that "similar individuals should be treated similarly". A primary difficulty with this definition is that it assumes a completely specified fairness metric for the task at hand. In contrast, we consider a framework for fairness elicitation, in which fairness is indirectly specified only via a sample of pairs of individuals who should be treated (approximately) equally on the task. We make no assumption that these pairs are consistent with any metric. We provide a provably convergent oracle-efficient algorithm for minimizing error subject to the fairness constraints, and prove generalization theorems for both accuracy and fairness. Since the constrained pairs could be elicited either from a panel of judges, or from particular individuals, our framework provides a means for algorithmically enforcing subjective notions of fairness. We report on preliminary findings of a behavioral study of subjective fairness using human-subject fairness constraints elicited on the COMPAS criminal recidivism dataset.

Random forests remain among the most popular off-the-shelf supervised learning algorithms. Despite their well-documented empirical success, however, until recently, few theoretical results were available to describe their performance and behavior. In this work we push beyond recent work on consistency and asymptotic normality by establishing rates of convergence for random forests and other supervised learning ensembles. We develop the notion of generalized U-statistics and show that within this framework, random forest predictions remain asymptotically normal for larger subsample sizes than previously established. We also provide Berry-Esseen bounds in order to quantify the rate at which this convergence occurs, making explicit the roles of the subsample size and the number of trees in determining the distribution of random forest predictions.

Previous work shows that adversarially robust generalization requires larger sample complexity, and the same dataset, e.g., CIFAR-10, which enables good standard accuracy may not suffice to train robust models. Since collecting new training data could be costly, we instead focus on inducing locally dense sample distribution, i.e., high sample density in the feature space which could lead to locally sufficient samples for robust learning. We first formally show that the softmax cross-entropy (SCE) loss and its variants induce inappropriate sample density distributions in the feature space, which inspires us to design appropriate training objectives. Specifically, we propose the Max-Mahalanobis center (MMC) loss to create high-density regions for better robustness. It encourages the learned features to gather around the preset class centers with optimal inter-class dispersion. Comparing to the SCE loss and its variants, we empirically demonstrate that applying the MMC loss can significantly improve robustness even under strong adaptive attacks, while keeping state-of-the-art accuracy on clean inputs with little extra computation.

Graph convolutional networks (GCNs) are powerful tools for graph-structured data. However, they have been recently shown to be prone to topological attacks. Despite substantial efforts to search for new architectures, it still remains a challenge to improve performance in both benign and adversarial situations simultaneously. In this paper, we re-examine the fundamental building block of GCN---the Laplacian operator---and highlight some basic flaws in the spatial and spectral domains. As an alternative, we propose an operator based on graph powering, and prove that it enjoys a desirable property of "spectral separation." Based on the operator, we propose a robust learning paradigm, where the network is trained on a family of "'smoothed" graphs that span a spatial and spectral range for generalizability. We also use the new operator in replacement of the classical Laplacian to construct an architecture with improved spectral robustness, expressivity and interpretability. The enhanced performance and robustness are demonstrated in extensive experiments.