Advances in scaling deep learning algorithms

Les algorithmes d'apprentissage profond forment un nouvel ensemble de méthodes puissantes pour l'apprentissage automatique. L'idée est de combiner des couches de facteurs latents en hierarchies. Cela requiert souvent un coût computationel plus elevé et augmente aussi le nombre de paramètres du modèle. Ainsi, l'utilisation de ces méthodes sur des problèmes à plus grande échelle demande de réduire leur coût et aussi d'améliorer leur régularisation et leur optimization. Cette thèse adresse cette question sur ces trois perspectives. Nous étudions tout d'abord le problème de réduire le coût de certains algorithmes profonds. Nous proposons deux méthodes pour entrainer des machines de Boltzmann restreintes et des auto-encodeurs débruitants sur des distributions sparses à haute dimension. Ceci est important pour l'application de ces algorithmes pour le traitement de langues naturelles. Ces deux méthodes (Dauphin et al., 2011; Dauphin and Bengio, 2013) utilisent l'échantillonage par importance pour échantilloner l'objectif de ces modèles. Nous observons que cela réduit significativement le temps d'entrainement. L'accéleration atteint 2 ordres de magnitude sur plusieurs bancs d'essai. Deuxièmement, nous introduisont un puissant régularisateur pour les méthodes profondes. Les résultats expérimentaux démontrent qu'un bon régularisateur est crucial pour obtenir de bonnes performances avec des gros réseaux (Hinton et al., 2012). Dans Rifai et al. (2011), nous proposons un nouveau régularisateur qui combine l'apprentissage non-supervisé et la propagation de tangente (Simard et al., 1992). Cette méthode exploite des principes géometriques et permit au moment de la publication d'atteindre des résultats à l'état de l'art. Finalement, nous considérons le problème d'optimiser des surfaces non-convexes à haute dimensionalité comme celle des réseaux de neurones. Tradionellement, l'abondance de minimum locaux était considéré comme la principale difficulté dans ces problèmes. Dans Dauphin et al. (2014a) nous argumentons à partir de résultats en statistique physique, de la théorie des matrices aléatoires, de la théorie des réseaux de neurones et à partir de résultats expérimentaux qu'une difficulté plus profonde provient de la prolifération de points-selle. Dans ce papier nous proposons aussi une nouvelle méthode pour l'optimisation non-convexe.

Neural changes associated with semantic processing in healthy aging despite intact behavioral performance

Semantic memory recruits an extensive neural network including the left inferior prefrontal cortex (IPC) and the left temporoparietal region, which are involved in semantic control processes, as well as the anterior temporal lobe region (ATL) which is considered to be involved in processing semantic information at a central level. However, little is known about the underlying neuronal integrity of the semantic network in normal aging. Young and older healthy adults carried out a semantic judgment task while their cortical activity was recorded using magnetoencephalography (MEG). Despite equivalent behavioral performance, young adults activated the left IPC to a greater extent than older adults, while the latter group recruited the temporoparietal region bilaterally and the left ATL to a greater extent than younger adults. Results indicate that significant neuronal changes occur in normal aging, mainly in regions underlying semantic control processes, despite an apparent stability in performance at the behavioral level.

Nutrient Dynamics in the EEZ of the West Coast of India with Special Reference to the Oxygen Minimum Zone and Denitrification

The present work aims at deciphering the processes that control the nutrient distribution along the EEZ (Exclusive Economic Zone of India) of the west coast of India and to bring out its linkage with primary and secondary productivity. This work assume utmost importance as very few studies have hitherto focused entirely on the EEZ of the west coast of India to address the biochemical responses brought about by monsoons. The present study examines the seasonal variations in physicochemical parameters and associated primary biological responses along the west coast of India. This study targets to measure and understand the shelf ocean exchange in a typical coastal upwelling region of the southeast Arabian Sea, and the influence of convective mixing along the northern part of the west coast of India. The study focuses more directly on coastal upwelling along the southwest coast of India, within the EEZ. The effects of coastal upwelling, eddy formation and the offshore advection are apparent in the present investigation. This has consequences to fisheries and climate, in energy transfer to the food chain and the increased sequestering of carbon in the ocean. The study also focuses on the Oxygen Minimum Zone (OMZ) and dentrification observed along the EEZ of the west coast of India on a seasonal scale. In the study, an attempt is also made to demarcate the geographical boundaries of the denitrification zone in the EEZ of India and on the nature and magnitude of these variations, on a seasonal and inter annual scales

Effect of processing parameters on the mechanical properties of short Kevlar aramid fibre-thermoplastic polyurethane composite

The effect of various processing parameters, such as nip gap, friction ratio and roll temperature, on the tensile properties of short Kevlar aramid fibre-thermoplastic polyurethane composite has been investigated and the tensile and tear fracture surfaces have been characterised using a scanning electron microscope. A nip gap of 0.45 mm, a friction ratio of 1.15 and a roll temperature of 62°C was found to give optimum mechanical properties. Scanning electron microscopy study revealed a higher extent of fibre orientation in the milling direction in the above condition.

Studies on the curing of short polyester fiber-polyurethane elastomer composite with different interfacial bonding agents

The cure characteristics of short fiber-polyurethane elastomer were studied with respect to different fiber-matrix bonding agents. A hexamethylenetetramine- resorcinol -hydrated silica based bonding agent was found to affect the stability of the composite. A new bonding agent, TP resin, based on polymeric toluenediisocyanate and polypropylene glycol has been developed. Cure characteristics of the composite with and without TP resin at different fiber loadings were also compared. Minimum torque, scorch time and optimum cure time increased with fiber content. Maximum torque was consistently higher with TP resin at all fiber loadings.

Production and Characterization of Endocylanase from Bacillus Pumilus Using Solidstate Fermentation and its Application in 'Paper Pulp Processing

In the present studies it is clear that Bacillus pumilus xylanase is having the characteristic suited for an industrial enzyme (xylanases that are active and stable at elevated temperatures and alkaline pH are needed). SSF production of xylanases and its application appears to be an innovative technology where the fermented substrate is the enzyme source that is used directly in the bleaching process without a prior downstream processing. The direct use of SSF enzymes in bleaching is a relatively new biobleaching approach. This can certainly benefit the bleaching process to lower the xylanase production costs and improve the economics and viability of the biobleaching technology. The application of enzymes to the bleaching process has been considered as an environmentally friendly approach that can reduce the negative impact on the environment exerted by the use of chlorine-based bleaching agents. It has been demonstrated that pretreatment of kraft pulp with xylanase prior to bleaching (biobleaching) can facilitate subsequent removal of lignin by bleaching chemicals, thereby, reducing the demand for elemental chlorine or improving final paper brightness. Using this xylanase pre-treatment, has resulted in an increased of brightness (8.5 Unit) when compared to non-enzymatic treated bleached pulp prepared using identical conditions. Reduction of the consumption of active chlorine can be achieved which results in a decrease in the toxicity, colour, chloride and absorbable organic halogen (AOX) levels of bleaching effluents. The xylanase treatment improves drainage, strength properties and the fragility of pulps, and also increases the brightness of pulps. This positive result shows that enzyme pre-treatment facilitates the removal of chromophore fragments of pulp there by making the process more environment friendly

Design and Development of an Adaptable Frame -Based System for Dravidian Language Processing

This work is aimed at building an adaptable frame-based system for processing Dravidian languages. There are about 17 languages in this family and they are spoken by the people of South India.Karaka relations are one of the most important features of Indian languages. They are the semabtuco-syntactic relations between verbs and other related constituents in a sentence. The karaka relations and surface case endings are analyzed for meaning extraction. This approach is comparable with the borad class of case based grammars.The efficiency of this approach is put into test in two applications. One is machine translation and the other is a natural language interface (NLI) for information retrieval from databases. The system mainly consists of a morphological analyzer, local word grouper, a parser for the source language and a sentence generator for the target language. This work make contributios like, it gives an elegant account of the relation between vibhakthi and karaka roles in Dravidian languages. This mapping is elegant and compact. The same basic thing also explains simple and complex sentence in these languages. This suggests that the solution is not just ad hoc but has a deeper underlying unity. This methodology could be extended to other free word order languages. Since the frame designed for meaning representation is general, they are adaptable to other languages coming in this group and to other applications.

Influence of Processing Variables on Protein Quality and Frozen Storage Stability of Two Commercially Important Species of Squid (Loligo duvaucelii and Doryteuthis sibogae )

Cephalopods are utilized as an important food item in various countries because of its delicacy as raw consumed food. Mainly sepia and loligo are consumed raw by Japanese and Russians. The freshness of the products is very important when the product is consumed raw. The major species that dominate our squid catch are Loligo duvaucelii and Doryteuthis sibogae. There is a noticeable difference in the quality of both the species. The needle squid (Doryteuthis sibogae ) contributes about 35% of the total squid landing. Due to the fast deterioration , a major portion of the needle squid, which is caught during the first few hauls, is thrown back to sea. The catch in the last hauls only are taken to the landing centers. At present the needle squid is processed as blanched rings and the desired quality is not obtained if it is processed as whole, whole cleaned or as tubes. In this study an attempt is made to investigate the biochemical characteristics in both the species of squid in relation to their quality and, the process control measures to be adopted. The effect of various treatments on their quality and the changes in proteolytic and lysosomal enzymes under various processing conditions are also studied in detail.Thus this study can provide the seafood industry with relevant suggestions and solutions for effective utilization of both the species of squid with emphasis on needle squid.

A Multilingual Query Processing System using Software Agents

The goal of this work was developing a query processing system using software agents. Open Agent Architecture framework is used for system development. The system supports queries in both Hindi and Malayalam; two prominent regional languages of India. Natural language processing techniques are used for meaning extraction from the plain query and information from database is given back to the user in his native language. The system architecture is designed in a structured way that it can be adapted to other regional languages of India. . This system can be effectively used in application areas like e-governance, agriculture, rural health, education, national resource planning, disaster management, information kiosks etc where people from all walks of life are involved.