Principal component analysis (PCA) is one of the most widely used dimension reduction and multivariate statistical techniques. From a probabilistic perspective, PCA seeks a low-dimensional representation of data in the presence of independent identical Gaussian noise. Probabilistic PCA (PPCA) and its variants have been extensively studied for decades. Most of them assume the underlying noise follows a certain independent identical distribution. However, the noise in the real world is usually complicated and structured. To address this challenge, some non-linear variants of PPCA have been proposed. But those methods are generally difficult to interpret. To this end, we propose a powerful and intuitive PCA method (MN-PCA) through modeling the graphical noise by the matrix normal distribution, which enables us to explore the structure of noise in both the feature space and the sample space. MN-PCA obtains a low-rank representation of data and the structure of noise simultaneously. And it can be explained as approximating data over the generalized Mahalanobis distance. We develop two algorithms to solve this model: one maximizes the regularized likelihood, the other exploits the Wasserstein distance, which is more robust. Extensive experiments on various data demonstrate their effectiveness.

This study investigates a new constrained-optimization formulation for deep ordinal classification. We impose uni-modality of the label distribution implicitly via a set of inequality constraints over pairs of adjacent labels. To tackle the ensuing challenging optimization problem, we solve a sequence of unconstrained losses based on a powerful extension of the log-barrier method. This accommodates standard SGD for deep networks, and avoids computationally expensive Lagrangian dual steps and projections, while outperforming substantially penalty methods. Our non-parametric model is more flexible than the existing deep ordinal classification techniques: it does not restrict the learned representation to a specific parametric model, allowing the training to explore larger spaces of solutions and removing the need for ad hoc choices, while scaling up to large numbers of labels. It can be used in conjunction with any standard classification loss and any deep architecture. We also propose a new performance metric for ordinal classification, as a proxy to measure a distribution uni-modality, referred to as the Sides Order Index (SOI). We report comprehensive evaluations and comparisons to state-of-the-art methods on benchmark public datasets for several ordinal classification tasks, showing the merits of our approach in terms of label consistency and scalability. A public reproducible PyTorch implementation is provided (https://github.com/sbelharbi/Deep-Ordinal-Classification-with-Inequality-Constraints).

Within the Private Equity (PE) market, the event of a private company undertaking an Initial Public Offering (IPO) is usually a very high-return one for the investors in the company. For this reason, an effective predictive model for the IPO event is considered as a valuable tool in the PE market, an endeavor in which publicly available quantitative information is generally scarce. In this paper, we describe a data-analytic procedure for predicting the probability with which a company will go public in a given forward period of time. The proposed method is based on the interplay of a neural network (NN) model for estimating the overall event probability, and Survival Analysis (SA) for further modeling the probability of the IPO event in any given interval of time. The proposed neuro-survival model is tuned and tested across nine industrial sectors using real data from the Thomson Reuters Eikon PE database.

In this paper we propose a new augmentation technique, called patch augmentation, that, in our experiments, improves model accuracy and makes networks more robust to adversarial attacks. In brief, this data-independent approach creates new image data based on image/label pairs, where a patch from one of the two images in the pair is superimposed on to the other image, creating a new augmented sample. The new image's label is a linear combination of the image pair's corresponding labels. Initial experiments show a several percentage point increase in accuracy on CIFAR-10, from a baseline of approximately 81% to 89%. CIFAR-100 sees larger improvements still, from a baseline of 52% to 68% accuracy. Networks trained using patch augmentation are also more robust to adversarial attacks, which we demonstrate using the Fast Gradient Sign Method.

The nearest-centroid classifier is a simple linear-time classifier based on computing the centroids of the data classes in the training phase, and then assigning a new datum to the class corresponding to its nearest centroid. Thanks to its very low computational cost, the nearest-centroid classifier is still widely used in machine learning, despite the development of many other more sophisticated classification methods. In this paper, we propose two sparse variants of the nearest-centroid classifier, based respectively on $\ell_1$ and $\ell_2$ distance criteria. The proposed sparse classifiers perform simultaneous classification and feature selection, by detecting the features that are most relevant for the classification purpose. We show that training of the proposed sparse models, with both distance criteria, can be performed exactly (i.e., the globally optimal set of features is selected) and at a quasi-linear computational cost. The experimental results show that the proposed methods are competitive in accuracy with state-of-the-art feature selection techniques, while having a significantly lower computational cost.

W e explore the problem of selectively forgetting a particular subset of the data used for training a deep neural network. While the effects of the data to be forgotten can be hidden from the output of the network, insights may still be gleaned by probing deep into its weights. W e propose a method for "scrubbing" the weights clean of information about a particular set of training data. The method does not require retraining from scratch, nor access to the data originally used for training. Instead, the weights are modified so that any probing function of the weights, computed with no knowledge of the random seed used for training, is indistinguishable from the same function applied to the weights of a network trained without the data to be forgotten. This condition is a generalized and weaker form of Differential Privacy. Exploiting ideas related to the stability of stochastic gradient descent, we introduce an upper-bound on the amount of information remaining in the weights, which can be estimated efficiently even for deep neural networks.

The high cost of communicating gradients is a major bottleneck for federated learning, as the bandwidth of the participating user devices is limited. Existing gradient compression algorithms are mainly designed for data centers with high-speed network and achieve $O(\sqrt{d} \log d)$ per-iteration communication cost at best, where $d$ is the size of the model. We propose hyper-sphere quantization (HSQ), a general framework that can be configured to achieve a continuum of trade-offs between communication efficiency and gradient accuracy. In particular, at the high compression ratio end, HSQ provides a low per-iteration communication cost of $O(\log d)$, which is favorable for federated learning. We prove the convergence of HSQ theoretically and show by experiments that HSQ significantly reduces the communication cost of model training without hurting convergence accuracy.

We explore adversarial robustness in the setting in which it is acceptable for a classifier to abstain---that is, output no class---on adversarial examples. Adversarial examples are small perturbations of normal inputs to a classifier that cause the classifier to give incorrect output; they present security and safety challenges for machine learning systems. In many safety-critical applications, it is less costly for a classifier to abstain on adversarial examples than to give incorrect output for them. We first introduce a novel objective function for adversarial robustness with an abstain option which characterizes an explicit tradeoff between robustness and accuracy. We then present a simple baseline in which an adversarially-trained classifier abstains on all inputs within a certain distance of the decision boundary, which we theoretically and experimentally evaluate. Finally, we propose Combined Abstention Robustness Learning (CARL), a method for jointly learning a classifier and the region of the input space on which it should abstain. We explore different variations of the PGD and DeepFool adversarial attacks on CARL in the abstain setting. Evaluating against these attacks, we demonstrate that training with CARL results in a more accurate, robust, and efficient classifier than the baseline.

Graph kernels are widely used for measuring the similarity between graphs. Many existing graph kernels, which focus on local patterns within graphs rather than their global properties, suffer from significant structure information loss when representing graphs. Some recent global graph kernels, which utilizes the alignment of geometric node embeddings of graphs, yield state-of-the-art performance. However, these graph kernels are not necessarily positive-definite. More importantly, computing the graph kernel matrix will have at least quadratic {time} complexity in terms of the number and the size of the graphs. In this paper, we propose a new family of global alignment graph kernels, which take into account the global properties of graphs by using geometric node embeddings and an associated node transportation based on earth mover's distance. Compared to existing global kernels, the proposed kernel is positive-definite. Our graph kernel is obtained by defining a distribution over \emph{random graphs}, which can naturally yield random feature approximations. The random feature approximations lead to our graph embeddings, which is named as "random graph embeddings" (RGE). In particular, RGE is shown to achieve \emph{(quasi-)linear scalability} with respect to the number and the size of the graphs. The experimental results on nine benchmark datasets demonstrate that RGE outperforms or matches twelve state-of-the-art graph classification algorithms.

In this paper a new formulation of event recognition task is examined: it is required to predict event categories in a gallery of images, for which albums (groups of photos corresponding to a single event) are unknown. We propose the novel two-stage approach. At first, features are extracted in each photo using the pre-trained convolutional neural network. These features are classified individually. The scores of the classifier are used to group sequential photos into several clusters. Finally, the features of photos in each group are aggregated into a single descriptor using neural attention mechanism. This algorithm is optionally extended to improve the accuracy for classification of each image in an album. In contrast to conventional fine-tuning of convolutional neural networks (CNN) we proposed to use image captioning, i.e., generative model that converts images to textual descriptions. They are one-hot encoded and summarized into sparse feature vector suitable for learning of arbitrary classifier. Experimental study with Photo Event Collection and Multi-Label Curation of Flickr Events Dataset demonstrates that our approach is 9-20% more accurate than event recognition on single photos. Moreover, proposed method has 13-16% lower error rate than classification of groups of photos obtained with hierarchical clustering. It is experimentally shown that the image captions trained on Conceptual Captions dataset can be classified more accurately than the features from object detector, though they both are obviously not as rich as the CNN-based features. However, it is possible to combine our approach with conventional CNNs in an ensemble to provide the state-of-the-art results for several event datasets.