Université de Montréal

About: Université de Montréal is a education organization based out in Montreal, Quebec, Canada. It is known for research contribution in the topics: Population & Context (language use). The organization has 45641 authors who have published 100476 publications receiving 4004007 citations. The organization is also known as: University of Montreal & UdeM.

How to Construct Deep Recurrent Neural Networks

Abstract: In this paper, we explore different ways to extend a recurrent neural network (RNN) to a \textit{deep} RNN. We start by arguing that the concept of depth in an RNN is not as clear as it is in feedforward neural networks. By carefully analyzing and understanding the architecture of an RNN, however, we find three points of an RNN which may be made deeper; (1) input-to-hidden function, (2) hidden-to-hidden transition and (3) hidden-to-output function. Based on this observation, we propose two novel architectures of a deep RNN which are orthogonal to an earlier attempt of stacking multiple recurrent layers to build a deep RNN (Schmidhuber, 1992; El Hihi and Bengio, 1996). We provide an alternative interpretation of these deep RNNs using a novel framework based on neural operators. The proposed deep RNNs are empirically evaluated on the tasks of polyphonic music prediction and language modeling. The experimental result supports our claim that the proposed deep RNNs benefit from the depth and outperform the conventional, shallow RNNs.

Disturbed dreaming, posttraumatic stress disorder, and affect distress: A review and neurocognitive model.

Abstract: Nightmares are common, occurring weekly in 4%–10% of the population, and are associated with female gender, younger age, increased stress, psychopathology, and dispositional traits. Nightmare pathogenesis remains unexplained, as do differences between nontraumatic and posttraumatic nightmares (for those with or without posttraumatic stress disorder) and relations with waking functioning. No models adequately explain nightmares nor have they been reconciled with recent developments in cognitive neuroscience, fear acquisition, and emotional memory. The authors review the recent literature and propose a conceptual framework for understanding a spectrum of dysphoric dreaming. Central to this is the notion that variations in nightmare prevalence, frequency, severity, and psychopathological comorbidity reflect the influence of both affect load, a consequence of daily variations in emotional pressure, and affect distress, a disposition to experience events with distressing, highly reactive emotions. In a cross-state, multilevel model of dream function and nightmare production, the authors integrate findings on emotional memory structures and the brain correlates of emotion.

Common variants in the NLRP3 region contribute to Crohn's disease susceptibility

Abstract: We used a candidate gene approach to identify a set of SNPs, located in a predicted regulatory region on chromosome 1q44 downstream of NLRP3 (previously known as CIAS1 and NALP3) that are associated with Crohn's disease. The associations were consistently replicated in four sample sets from individuals of European descent. In the combined analysis of all samples (710 father-mother-child trios, 239 cases and 107 controls), these SNPs were strongly associated with risk of Crohn's disease (P(combined) = 3.49 x 10(-9), odds ratio = 1.78, confidence interval = 1.47-2.16 for rs10733113), reaching a level consistent with the stringent significance thresholds imposed by whole-genome association studies. In addition, we observed significant associations between SNPs in the associated regions and NLRP3 expression and IL-1beta production. Mutations in NLRP3 are known to be responsible for three rare autoinflammatory disorders. These results suggest that the NLRP3 region is also implicated in the susceptibility of more common inflammatory diseases such as Crohn's disease.

Convergence Properties of the K-Means Algorithms

Abstract: This paper studies the convergence properties of the well known K-Means clustering algorithm. The K-Means algorithm can be described either as a gradient descent algorithm or by slightly extending the mathematics of the EM algorithm to this hard threshold case. We show that the K-Means algorithm actually minimizes the quantization error using the very fast Newton algorithm.