Given a state-of-the-art deep neural network text classifier, we show the existence of a universal and very small perturbation vector (in the embedding space) that causes natural text to be misclassified with high probability. Unlike images on which a single fixed-size adversarial perturbation can be found, text is of variable length, so we define the "universality" as "token-agnostic", where a single perturbation is applied to each token, resulting in different perturbations of flexible sizes at the sequence level. W e propose an algorithm to compute universal adversarial perturbations, and show that the state-of-the-art deep neural networks are highly vulnerable to them, even though they keep the neighborhood of tokens mostly preserved. W e also show how to use these adversarial perturbations to generate adversarial text samples. The surprising existence of universal "token-agnostic" adversarial perturbations may reveal important properties of a text classifier.

We consider the problem of manifold learning. Extending existing approaches of learning from randomly sampled data points, we consider contexts where data may be chosen by a teacher. We analyze learning from teachers who can provide structured data such as points, comparisons (pairs of points), demonstrations (sequences). We prove results showing that the former two do not yield notable decreases in the amount of data required to infer a manifold. Teaching by demonstration can yield remarkable decreases in the amount of data required, if we allow the goal to be teaching up to topology. We further analyze teaching learners in the context of persistence homology. Teaching topology can greatly reduce the number of datapoints required to infer correct geometry, and allows learning from teachers who themselves do not have full knowledge of the true manifold. We conclude with implications for learning in humans and machines.

Gaussian Processes (GPs) are powerful non-parametric Bayesian regression models that allow exact posterior inference, but exhibit high computational and memory costs. In order to improve scalability of GPs, approximate posterior inference is frequently employed, where a prominent class of approximation techniques is based on local GP experts. However, the local-expert techniques proposed so far are either not well-principled, come with limited approximation guarantees, or lead to intractable models. In this paper, we introduce deep structured mixtures of GP experts, a stochastic process model which i) allows exact posterior inference, ii) has attractive computational and memory costs, and iii), when used as GP approximation, captures predictive uncertainties consistently better than previous approximations. In a variety of experiments, we show that deep structured mixtures have a low approximation error and outperform existing expert-based approaches.

We propose probabilistic models that can extrapolate learning curves of iterative machine learning algorithms, such as stochastic gradient descent for training deep networks, based on training data with variable-length learning curves. We study instantiations of this framework based on random forests and Bayesian recurrent neural networks. Our experiments show that these models yield better predictions than state-of-the-art models from the hyperparameter optimization literature when extrapolating the performance of neural networks trained with different hyperparameter settings.

Online retail, eCommerce, frequently falls victim to fraud conducted by malicious customers (fraudsters) who obtain goods or services through deception. Fraud coordinated by groups of professional fraudsters that place several fraudulent orders to maximize their gain is referred to as organized fraud. Existing approaches to fraud detection typically analyze orders in isolation and they are not effective at identifying groups of fraudulent orders linked to organized fraud. These also wrongly identify many legitimate orders as fraud, which hinders their usage for automated fraud cancellation. We introduce a novel solution to detect organized fraud by analyzing orders in bulk. Our approach is based on clustering and aims to group together fraudulent orders placed by the same group of fraudsters. It selectively uses two existing techniques, agglomerative clustering and sampling to recursively group orders into small clusters in a reasonable amount of time. We assess our clustering technique on real-world orders placed on the Zalando website, the largest online apparel retailer in Europe1. Our clustering processes 100,000s of orders in a few hours and groups 35-45% of fraudulent orders together. We propose a simple technique built on top of our clustering that detects 26.2% of fraud while raising false alarms for only 0.1% of legitimate orders.

Multi-head attention mechanism is capable of learning various representations from sequential data while paying attention to different subsequences, e.g., word-pieces or syllables in a spoken word. From the subsequences, it retrieves richer information than a single-head attention which only summarizes the whole sequence into one context vector. However, a naive use of the multi-head attention does not guarantee such richness as the attention heads may have positional and representational redundancy. In this paper, we propose a regularization technique for multi-head attention mechanism in an end-to-end neural keyword spotting system. Augmenting regularization terms which penalize positional and contextual non-orthogonality between the attention heads encourages to output different representations from separate subsequences, which in turn enables leveraging structured information without explicit sequence models such as hidden Markov models. In addition, intra-head contextual non-orthogonality regularization encourages each attention head to have similar representations across keyword examples, which helps classification by reducing feature variability. The experimental results demonstrate that the proposed regularization technique significantly improves the keyword spotting performance for the keyword "Hey Snapdragon".

We study a variant of Wasserstein barycenter problem, which we refer to as \emph{tree-sliced Wasserstein barycenter}, by leveraging the structure of tree metrics for the ground metrics in the formulation of Wasserstein distance. Drawing on the tree structure, we propose efficient algorithms for solving the unconstrained and constrained versions of tree-sliced Wasserstein barycenter. The algorithms have fast computational time and efficient memory usage, especially for high dimensional settings while demonstrating favorable results when the tree metrics are appropriately constructed. Experimental results on large-scale synthetic and real datasets from Wasserstein barycenter for documents with word embedding, multilevel clustering, and scalable Bayes problems show the advantages of tree-sliced Wasserstein barycenter over (Sinkhorn) Wasserstein barycenter.

Contrastive unsupervised representation learning (CURL) is the state-of-the-art technique to learn representations (as a set of features) from unlabelled data. While CURL has collected several empirical successes recently, theoretical understanding of its performance was still missing. In a recent work, Arora et al. ( 2019) provide the first generalisation bounds for CURL, relying on a Rademacher complexity. We extend their framework to the flexible PAC-Bayes setting, allowing to deal with the non-iid setting. We present PAC-Bayesian generalisation bounds for CURL, which are then used to derive a new representation learning algorithm. Numerical experiments on real-life datasets illustrate that our algorithm achieves competitive accuracy, and yields generalisation bounds with non-vacuous values.

We propose two novel variants of Gromov-Wasserstein (GW) between probability measures in different probability spaces based on projecting these measures into the tree metric spaces. Our first proposed discrepancy, named \emph{flow-based tree Gromov-Wasserstein}, hinges upon the tree metric from node to root in each tree to define the structure representation of probability measures on trees. The flow-based tree GW shares similar structures with univariate Wasserstein distance while keeping sufficient spatial information of the original projected probability measures. In order to further explore the structure of tree, we proposed another version of flow-based tree GW, which we refer to as \emph{depth-based tree Gromov-Wasserstein}. That discrepancy considers the alignment of probability measures hierarchically along each depth level of the tree structures. Finally, we demonstrate via extensive simulation studies on large-scale real data sets the relative advantage of the proposed discrepancies.

"No free lunch" results state the impossibility of obtaining meaningful bounds on the error of a learning algorithm without prior assumptions and modelling. Some models are expensive (strong assumptions, such as as subgaussian tails), others are cheap (simply finite variance). As it is well known, the more you pay, the more you get: in other words, the most expensive models yield the more interesting bounds. Recent advances in robust statistics have investigated procedures to obtain tight bounds while keeping the cost minimal. The present paper explores and exhibits what the limits are for obtaining tight PAC-Bayes bounds in a robust setting for cheap models, addressing the question: is PAC-Bayes good value for money?