Designing Regularizers and Architectures for Recurrent Neural Networks

Cette thèse contribue a la recherche vers l'intelligence artificielle en utilisant des méthodes connexionnistes. Les réseaux de neurones récurrents sont un ensemble de modèles séquentiels de plus en plus populaires capable en principe d'apprendre des algorithmes arbitraires. Ces modèles effectuent un apprentissage en profondeur, un type d'apprentissage machine. Sa généralité et son succès empirique en font un sujet intéressant pour la recherche et un outil prometteur pour la création de l'intelligence artificielle plus générale. Le premier chapitre de cette thèse donne un bref aperçu des sujets de fonds: l'intelligence artificielle, l'apprentissage machine, l'apprentissage en profondeur et les réseaux de neurones récurrents. Les trois chapitres suivants couvrent ces sujets de manière de plus en plus spécifiques. Enfin, nous présentons quelques contributions apportées aux réseaux de neurones récurrents. Le chapitre \ref{arxiv1} présente nos travaux de régularisation des réseaux de neurones récurrents. La régularisation vise à améliorer la capacité de généralisation du modèle, et joue un role clé dans la performance de plusieurs applications des réseaux de neurones récurrents, en particulier en reconnaissance vocale. Notre approche donne l'état de l'art sur TIMIT, un benchmark standard pour cette tâche. Le chapitre \ref{cpgp} présente une seconde ligne de travail, toujours en cours, qui explore une nouvelle architecture pour les réseaux de neurones récurrents. Les réseaux de neurones récurrents maintiennent un état caché qui représente leurs observations antérieures. L'idée de ce travail est de coder certaines dynamiques abstraites dans l'état caché, donnant au réseau une manière naturelle d'encoder des tendances cohérentes de l'état de son environnement. Notre travail est fondé sur un modèle existant; nous décrivons ce travail et nos contributions avec notamment une expérience préliminaire.

The recurrent shortage of registered nurses : a new look at the issues.

Mode of access: Internet.

Recurrent maladies in scholarly writing.

Mode of access: Internet.

The Grand Canyon of the Colorado; recurrent studies in impressions and appearances,

Mode of access: Internet.

Powered microdebridement treatment for recurrent respiratory papillomatosis

Laryngeal papillomatosis is a benign disease of lhe larynx caused by human papilloma virus. The disease has u variable clinical course and treatment focuses on debridement until clinical remission. The most common technique for removing the papilloma is by carbon dioxide laser ublution. Powered microdebridement. which is more familiar to endoscopic sinus surgeons, has been adapted for use in the larynx. We would like to report on this technique for removal of respiratory papillomas that we believe to be safer for both patients and staff. The cases of seven paediatric patients with recurrent respiratory papillomatosis treated with microdebridement of their papillomas have been retrospectively reviewed.

CCR5-Delta 32 mutation is strongly associated with primary sclerosing cholangitis

CCR5 plays a key role in the distribution of CD45RO+ T cells and contributes to generation of a T helper 1 immune response. CCR5-Delta32 is a 32-bp deletion associated with significant reduction in cell surface expression of the receptor. We investigated the role of CCR5-Delta32 on susceptibility to ulcerative colitis (UC), Crohn's disease ( CD) and primary sclerosing cholangitis (PSC). Genotype and allelic association analyses were performed in 162 patients with UC, 131 with CD, 71 with PSC and 419 matched controls. There was a significant difference in CCR5 genotype (OR 2.27, P = 0.003) between patients with sclerosing cholangitis and controls. Similarly, CCR5-Delta32 allele frequency was significantly higher in sclerosing cholangitis (17.6%) compared to controls (9.9%, OR 2.47, P = 0.007) and inflammatory bowel disease patients without sclerosing cholangitis ( 11.3%, OR 1.9, P = 0.027). There were no significant differences in CCR5 genotype or allele frequency between those with either UC or CD and controls. Genotypes with the CCR5-Delta32 variant were increased in patients with severe liver disease defined by portal hypertension and/or transplantation (45%) compared to those with mild liver disease (21%, OR 3.17, P = 0.03). The CCR5-Delta32 mutation may influence disease susceptibility and severity in patients with PSC.

Generalization by symbolic abstraction in cascaded recurrent networks

Generalization performance in recurrent neural networks is enhanced by cascading several networks. By discretizing abstractions induced in one network, other networks can operate on a coarse symbolic level with increased performance on sparse and structural prediction tasks. The level of systematicity exhibited by the cascade of recurrent networks is assessed on the basis of three language domains. (C) 2004 Elsevier B.V. All rights reserved.

Management of recurrent epithelial ovarian carcinoma

Despite the standardisation of surgical techniques and significant progress in chemotherapeutics over the last 30 years, advanced epithelial ovarian cancer remains the most lethal gynaecological malignancy in the western world. Although the majority of women achieve a remission following primary therapy, most patients with advanced stage disease will eventually relapse and become candidates for 'salvage' therapy. The chances of a further remission depend on factors such as the 'treatment-free interval', and there are now a large number of chemotherapy agents with activity in ovarian cancer available to the oncologist. Recent randomised studies have reported on survival benefits for chemotherapy in recurrent disease, and therefore careful and appropriate selection of treatments has assumed a greater importance. This article reviews the most current data, and discusses the factors involved in making individualised treatment decisions.

Prediction of subcellular localization using sequence-biased recurrent networks

Motivation: Targeting peptides direct nascent proteins to their specific subcellular compartment. Knowledge of targeting signals enables informed drug design and reliable annotation of gene products. However, due to the low similarity of such sequences and the dynamical nature of the sorting process, the computational prediction of subcellular localization of proteins is challenging. Results: We contrast the use of feed forward models as employed by the popular TargetP/SignalP predictors with a sequence-biased recurrent network model. The models are evaluated in terms of performance at the residue level and at the sequence level, and demonstrate that recurrent networks improve the overall prediction performance. Compared to the original results reported for TargetP, an ensemble of the tested models increases the accuracy by 6 and 5% on non-plant and plant data, respectively.

The applicability of recurrent neural networks for biological sequence analysis

Selection of machine learning techniques requires a certain sensitivity to the requirements of the problem. In particular, the problem can be made more tractable by deliberately using algorithms that are biased toward solutions of the requisite kind. In this paper, we argue that recurrent neural networks have a natural bias toward a problem domain of which biological sequence analysis tasks are a subset. We use experiments with synthetic data to illustrate this bias. We then demonstrate that this bias can be exploitable using a data set of protein sequences containing several classes of subcellular localization targeting peptides. The results show that, compared with feed forward, recurrent neural networks will generally perform better on sequence analysis tasks. Furthermore, as the patterns within the sequence become more ambiguous, the choice of specific recurrent architecture becomes more critical.

Effects of a nurse-led, clinic and home-based intervention on recurrent hospital use in chronic heart failure

Background: Few studies have examined the potential benefits of specialist nurse-led programs of care involving home and clinic-based follow-up to optimise the post-discharge management of chronic heart failure (CHF). Objective: To determine the effectiveness of a hybrid program of clinic plus home-based intervention (C+HBI) in reducing recurrent hospitalisation in CHF patients. Methods: CHF patients with evidence of left ventricular systolic dysfunction admitted to two hospitals in Northern England were assigned to a C+HBI lasting 6 months post-discharge (n=58) or to usual, post-discharge care (UC: n=48) via a cluster randomization protocol. The co-primary endpoints were death or unplanned readmission (event-free survival) and rate of recurrent, all-cause readmission within 6 months of hospital discharge. Results: During study follow-up, more UC patients had an unplanned readmission for any cause (44% vs. 22%: P=0.0191 OR 1.95 95% CI 1.10-3.48) whilst 7 (15%) versus 5 (9%) UC and C+HBI patients, respectively, died (P=NS). Overall, 15 (26%) C+HBI versus 21 (44%) UC patients experienced a primary endpoint. C+HBI was associated with a non-significant, 45% reduction in the risk of death or readmission when adjusting for potential confounders (RR 0.55, 95% CI 0.28-1.08: P=0.08). Overall, C+HBI patients accumulated significantly fewer unplanned readmissions (15 vs. 45: P