Payment card fraud causes multibillion dollar losses for banks and merchants worldwide, often fueling complex criminal activities. To address this, many real-time fraud detection systems use tree-based models, demanding complex feature engineering systems to efficiently enrich transactions with historical data while complying with millisecond-level latencies. In this work, we do not require those expensive features by using recurrent neural networks and treating payments as an interleaved sequence, where the history of each card is an unbounded, irregular sub-sequence. We present a complete RNN framework to detect fraud in real-time, proposing an efficient ML pipeline from preprocessing to deployment. We show that these feature-free, multi-sequence RNNs outperform state-of-the-art models saving millions of dollars in fraud detection and using fewer computational resources.

We propose a joint formulation for learning task-specific cross-lingual word embeddings, along with classifiers for that task.Unlike prior work, which first learns the embeddings from parallel data and then plugs them in a supervised learning problem, our approach is oneshot: a single optimization problem combines a co-regularizer for the multilingual embeddings with a task-specific loss.We present theoretical results showing the limitation of Euclidean co-regularizers to increase the embedding dimension, a limitation which does not exist for other co-regularizers (such as the ℓ 1distance).Despite its simplicity, our method achieves state-of-the-art accuracies on the RCV1/RCV2 dataset when transferring from English to German, with training times below 1 minute.On the TED Corpus, we obtain the highest reported scores on 10 out of 11 languages.

Mariana S. C. Almeida, Cláudia Pinto, Helena Figueira, Pedro Mendes, André F. T. Martins. Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2015.

We address the problem of automatically attributing quotations to speakers, which has great relevance in text mining and media monitoring applications.While current systems report high accuracies for this task, they either work at mentionlevel (getting credit for detecting uninformative mentions such as pronouns), or assume the coreferent mentions have been detected beforehand; the inaccuracies in this preprocessing step may lead to error propagation.In this paper, we introduce a joint model for entity-level quotation attribution and coreference resolution, exploiting correlations between the two tasks.We design an evaluation metric for attribution that captures all speakers' mentions.We present results showing that both tasks benefit from being treated jointly.

The alternating direction method of multipliers (ADMM) is an efficient optimization tool that achieves state-of-the-art speed in several imaging inverse problems, by splitting the underlying problem into simpler, efficiently solvable sub-problems. In deconvolution, one of these sub-problems requires a matrix inversion, which has been shown to be efficiently computable (via the FFT), if the observation operator is circulant, i.e., under periodic boundary conditions. We extend ADMM-based image deconvolution to a more realistic scenario: unknown boundaries. The observation is modeled as the composition of a periodic convolution with a spatial mask that excludes the regions where the periodic convolution is invalid. We show that the resulting algorithms inherit the convergence guarantees of ADMM and illustrate its performance on non-periodic de-blurring under frame-based regularization.

Blind image deblurring (BID) is an ill-posed inverse problem, typically solved by imposing some form of regularization (prior knowledge) on the unknown blur and original image. A recent approach, although not requiring prior knowledge on the blurring filter, achieves state-of-the-art performance for a wide range of real-world BID problems. We propose a new version of that method, in which both the optimization problems with respect to the unknown image and with respect to the unknown blur are solved by the alternating direction method of multipliers (ADMM) - an optimization tool that has recently sparked much interest for solving inverse problems, namely due to its modularity and state-of-the-art speed. Our approach also handles seamlessly the realistic case of blind deblurring with unknown boundary conditions. Experiments with synthetic and real blurred images show the competitiveness of the proposed method, both in terms of speed and restoration quality.

Image deblurring (ID) is an ill-posed problem typically addressed by using regularization, or prior knowledge, on the unknown image (and also on the blur operator, in the blind case). ID is often formulated as an optimization problem, where the objective function includes a data term encouraging the estimated image (and blur, in blind ID) to explain the observed data well (typically, the squared norm of a residual) plus a regularizer that penalizes solutions deemed undesirable. The performance of this approach depends critically (among other things) on the relative weight of the regularizer (the regularization parameter) and on the number of iterations of the algorithm used to address the optimization problem. In this paper, we propose new criteria for adjusting the regularization parameter and/or the number of iterations of ID algorithms. The rationale is that if the recovered image (and blur, in blind ID) is well estimated, the residual image is spectrally white; contrarily, a poorly deblurred image typically exhibits structured artifacts (e.g., ringing, oversmoothness), yielding residuals that are not spectrally white. The proposed criterion is particularly well suited to a recent blind ID algorithm that uses continuation, i.e., slowly decreases the regularization parameter along the iterations; in this case, choosing this parameter and deciding when to stop are one and the same thing. Our experiments show that the proposed whiteness-based criteria yield improvements in SNR, on average, only 0.15 dB below those obtained by (clairvoyantly) stopping the algorithm at the best SNR. We also illustrate the proposed criteria on non-blind ID, reporting results that are competitive with state-of-the-art criteria (such as Monte Carlo-based GSURE and projected SURE), which, however, are not applicable for blind ID.

The alternating direction method of multipliers (ADMM) has recently sparked interest as a flexible and efficient optimization tool for inverse problems, namely, image deconvolution and reconstruction under non-smooth convex regularization. ADMM achieves state-of-the-art speed by adopting a divide and conquer strategy, wherein a hard problem is split into simpler, efficiently solvable sub-problems (e.g., using fast Fourier or wavelet transforms, or simple proximity operators). In deconvolution, one of these sub-problems involves a matrix inversion (i.e., solving a linear system), which can be done efficiently (in the discrete Fourier domain) if the observation operator is circulant, i.e., under periodic boundary conditions. This paper extends ADMM-based image deconvolution to the more realistic scenario of unknown boundary, where the observation operator is modeled as the composition of a convolution (with arbitrary boundary conditions) with a spatial mask that keeps only pixels that do not depend on the unknown boundary. The proposed approach also handles, at no extra cost, problems that combine the recovery of missing pixels (i.e., inpainting) with deconvolution. We show that the resulting algorithms inherit the convergence guarantees of ADMM and illustrate its performance on non-periodic deblurring (with and without inpainting of interior pixels) under total-variation and frame-based regularization.

A method for blind image deblurring is presented. The method only makes weak assumptions about the blurring filter and is able to undo a wide variety of blurring degradations. To overcome the ill-posedness of the blind image deblurring problem, the method includes a learning technique which initially focuses on the main edges of the image and gradually takes details into account. A new image prior, which includes a new edge detector, is used. The method is able to handle unconstrained blurs, but also allows the use of constraints or of prior information on the blurring filter, as well as the use of filters defined in a parametric manner. Furthermore, it works in both single-frame and multiframe scenarios. The use of constrained blur models appropriate to the problem at hand, and/or of multiframe scenarios, generally improves the deblurring results. Tests performed on monochrome and color images, with various synthetic and real-life degradations, without and with noise, in single-frame and multiframe scenarios, showed good results, both in subjective terms and in terms of the increase of signal to noise ratio (ISNR) measure. In comparisons with other state of the art methods, our method yields better results, and shows to be applicable to a much wider range of blurs.

This paper presents a method for deblurring an image consisting of two layers (a foreground layer and a background layer) which have suffered different, unknown blurs. This is a situation of practical interest. For example, it is common to find images in which we have a foreground object (e.g. a car) which has motion blur while the background is sharp (or vice-versa), or in which a foreground object and the background have different out-of-focus blurs. We develop a model for this foreground + background degradation, and extend a previously introduced blind deblurring method to deal with this situation. As in the original blind deblurring method, the method presented here does not impose any strong constraints on the blurring filters. The method is almost completely blind, requiring, form the user, just a coarse indication of which are the foreground and background areas of the image. The method has been tested with synthetic degradations and with real-life photos. We present some of the results. In all the experiments, the method was able to reasonably recover, from single degraded images: the complete deblurred image, the deblurred foreground and background images, and a mask providing the segmentation between foreground and background.

A new method to perform blind image deblurring is proposed. Very few assumptions are made on the blurring filter and on the original image: the blurring filter is assumed to have limited support and the original image is assumed to be a sharp natural image. A new prior is used, which gives higher probability to images with sharp edges. The estimation of both the deblurred image and the blurring filter is made in a progressive way, first taking into account the main features of the image, and then proceeding to smaller details. The results obtained with synthetically blurred images are good, even when the blur operator is rather ill-conditioned and the blurred image is noisy. The method also yields improvements in real-life photographs with focus and motion blurs.