Sentence Compression with Supervised Machine Learning

The subject of the thesis is automatic sentence compression with machine learning, so that the compressed sentences remain both grammatical and retain their essential meaning. There are multiple possible uses for the compression of natural language sentences. In this thesis the focus is generation of television program subtitles, which often are compressed version of the original script of the program. The main part of the thesis consists of machine learning experiments for automatic sentence compression using different approaches to the problem. The machine learning methods used for this work are linear-chain conditional random fields and support vector machines. Also we take a look which automatic text analysis methods provide useful features for the task. The data used for machine learning is supplied by Lingsoft Inc. and consists of subtitles in both compressed an uncompressed form. The models are compared to a baseline system and comparisons are made both automatically and also using human evaluation, because of the potentially subjective nature of the output. The best result is achieved using a CRF - sequence classification using a rich feature set. All text analysis methods help classification and most useful method is morphological analysis. Tutkielman aihe on suomenkielisten lauseiden automaattinen tiivistäminen koneellisesti, niin että lyhennetyt lauseet säilyttävät olennaisen informaationsa ja pysyvät kieliopillisina. Luonnollisen kielen lauseiden tiivistämiselle on monta käyttötarkoitusta, mutta tässä tutkielmassa aihetta lähestytään television ohjelmien tekstittämisen kautta, johon käytännössä kuuluu alkuperäisen tekstin lyhentäminen televisioruudulle paremmin sopivaksi. Tutkielmassa kokeillaan erilaisia koneoppimismenetelmiä tekstin automaatiseen lyhentämiseen ja tarkastellaan miten hyvin erilaiset luonnollisen kielen analyysimenetelmät tuottavat informaatiota, joka auttaa näitä menetelmiä lyhentämään lauseita. Lisäksi tarkastellaan minkälainen lähestymistapa tuottaa parhaan lopputuloksen. Käytetyt koneoppimismenetelmät ovat tukivektorikone ja lineaarisen sekvenssin mallinen CRF. Koneoppimisen tukena käytetään tekstityksiä niiden eri käsittelyvaiheissa, jotka on saatu Lingsoft OY:ltä. Luotuja malleja vertaillaan Lopulta mallien lopputuloksia evaluoidaan automaattisesti ja koska teksti lopputuksena on jossain määrin subjektiivinen myös ihmisarviointiin perustuen. Vertailukohtana toimii kirjallisuudesta poimittu menetelmä. Tutkielman tuloksena paras lopputulos saadaan aikaan käyttäen CRF sekvenssi-luokittelijaa laajalla piirrejoukolla. Kaikki kokeillut teksin analyysimenetelmät auttavat luokittelussa, joista tärkeimmän panoksen antaa morfologinen analyysi.

Quadtree representation and compression of spatial data

Spatial data representation and compression has become a focus issue in computer graphics and image processing applications. Quadtrees, as one of hierarchical data structures, basing on the principle of recursive decomposition of space, always offer a compact and efficient representation of an image. For a given image, the choice of quadtree root node plays an important role in its quadtree representation and final data compression. The goal of this thesis is to present a heuristic algorithm for finding a root node of a region quadtree, which is able to reduce the number of leaf nodes when compared with the standard quadtree decomposition. The empirical results indicate that, this proposed algorithm has quadtree representation and data compression improvement when in comparison with the traditional method.

Compression de données d'animation acquises par capture de mouvements

Thèse numérisée par la Division de la gestion de documents et des archives de l'Université de Montréal

Development of Elastomeric Hybrid Composite Based on Synthesised Manosilica and Short Nylon Fiber

Nanoscale silica was synthesized by precipitation method using sodium silicate and dilute hydrochloric acid under controlled conditions. The synthesized silica was characterized by Scanning Electron Microscopy (SEM), Transmission Electron Microscopy (TEM), BET adsorption and X-Ray Diffraction (XRD). The particle size of silica was calculated to be 13 nm from the XRD results and the surface area was found to be 295 m2/g by BET method. The performance of this synthesized nanosilica as a reinforcing filler in natural rubber (NR) compound was investigated. The commercial silica was used as the reference material. Nanosilica was found to be effective reinforcing filler in natural rubber compound. Filler-matrix interaction was better for nanosilica than the commercial silica. The synthesized nanosilica was used in place of conventional silica in HRH (hexamethylene tetramine, resorcinol and silica) bonding system for natural rubber and styrene butadiene rubber / Nylon 6 short fiber composites. The efficiency of HRH bonding system based on nanosilica was better. Nanosilica was also used as reinforcing filler in rubber / Nylon 6 short fiber hybrid composite. The cure, mechanical, ageing, thermal and dynamic mechanical properties of nanosilica / Nylon 6 short fiber / elastomeric hybrid composites were studied in detail. The matrices used were natural rubber (NR), nitrile rubber (NBR), styrene butadiene rubber (SBR) and chloroprene rubber (CR). Fiber loading was varied from 0 to 30 parts per hundred rubber (phr) and silica loading was varied from 0 to 9 phr. Hexa:Resorcinol:Silica (HRH) ratio was maintained as 2:2:1. HRH loading was adjusted to 16% of the fiber loading. Minimum torque, maximum torque and cure time increased with silica loading. Cure rate increased with fiber loading and decreased with silica content. The hybrid composites showed improved mechanical properties in the presence of nanosilica. Tensile strength showed a dip at 10 phr fiber loading in the case of NR and CR while it continuously increased with fiber loading in the case of NBR and SBR. The nanosilica improved the tensile strength, modulus and tear strength better than the conventional silica. Abrasion resistance and hardness were also better for the nanosilica composites. Resilience and compression set were adversely affected. Hybrid composites showed anisotropy in mechanical properties. Retention in ageing improved with fiber loading and was better for nanosilica-filled hybrid composites. The nanosilica also improved the thermal stability of the hybrid composite better than the commercial silica. All the composites underwent two-step thermal degradation. Kinetic studies showed that the degradation of all the elastomeric composites followed a first-order reaction. Dynamic mechanical analysis revealed that storage modulus (E’) and loss modulus (E”) increased with nanosiica content, fiber loading and frequency for all the composites, independent of the matrix. The highest rate of increase was registered for NBR rubber.

Studies on the Utilisation of Rubber Reclaim in Elastomers

The thesis describes utilisation of reclaimed rubber, Whole Tyre Reclaim (WTR) produced from bio non- degradable solid pollutant scrap and used tyres. In this study an attempt has made to optimize the substitution of virgin rubber with WTR in both natural and synthetic rubber compounds without seriously compromising the important mechanical properties. The WTR is used as potent source of rubber hydrocarbon and carbon black filler. Apart from natural rubber (NR), Butadiene rubber (BR), Styrene butadiene rubber (SBR), Acrylonitrile butadiene rubber (NBR) and Chloroprene rubber (CR) were selected for study, being the most widely used general purpose and specialty rubbers. The compatibility problem was addressed by functionalisation of WTR with maleic anhydride and by using a coupling agent Si69.The blends were systematically evaluated with respect to various mechanical properties. The thermogravimetric analyses were also carried out to evaluate the thermal stability of the blends.Mechanical properties of the blends were property and matrix dependant. Presence of reinforcing carbon black filler and curatives in the reclaimed rubber improved the mechanical properties with the exception of some of the elastic properties like heat build up, resilience, compression set. When WTR was blended with natural rubber and synthetic rubbers, as the concentration of the low molecular weight, depolymerised WfR was increased above 46-weight percent, the properties deteriorates.When WTR was blended with crystallizing rubbers such as natural rubber and chloroprene rubber, properties like tensile strength, ultimate elongation were decreased in presence of WTR. Where as in the case of blends of WTR with non-crystallizing rubbers reinforcement effect was more prominent.The effect of functionalisation and coupling agent was studied in three matrices having different levels of polarity(NBR, CR and SBR).The grafting of maleic anhydride on to WTR definitely improved the properties of its blends with NBR, CR and SBR, the effect being prominent in Chloroprene rubber.Improvement in properties of these blends could also achieved by using a coupling agent Si69. With this there is apparent plasticizing effect at higher loading of the coupling agent. The optimum concentration of Si69 was 1 phr for improved properties, though the improvements are not as significant as in the case of maleic anhydride grafting.Thermal stability of the blend was increased by using silane-coupling agent.

Polymer Nanocomposites: Crystallization, Reinforcement and Conductivity through SWNTs

The current research investigates the possibility of using single walled carbon nanotubes (SWNTs) as filler in polymers to impart several properties to the matrix polymer. SWNTs in a polymer matrix like poly(ethylene terephthalate) induce nucleation in its melt crystallization, provide effective reinforcement and impart electrical conductivity. We adopt a simple melt compounding technique for incorporating the nanotubes into the polymer matrix. For attaining a better dispersion of the filler, an ultrasound assisted dissolution-evaporation method has also been tried. The resulting enhancement in the materials properties indicates an improved disentanglement of the nanotube ropes, which in turn provides effective matrix-filler interaction. PET-SWNT nanocomposite fibers prepared through melt spinning followed by subsequent drawing are also found to have significantly higher mechanical propertiesas compared to pristine PET fiber.SWNTs also find applications in composites based on elastomers such as natural rubber as they can impart electrical conductivity with simultaneous improvement in the mechanical properties.

Utilisation of Coir Geotextiles for Unpaved Roads and Embankments

The increasing tempo of construction activity the world over creates heavy pressure on existing land space. The quest for new and competent site often points to the needs for improving existing sites, which are otherwise deemed unsuitable for adopting conventional foundations. This is accomplished by ground improvement methods, which are employed to improve the quality of soil incompetent in their natural state. Among the construction activities, a well-connected road network is one of the basic infrastructure requirements, which play a vital role for the fast and comfortable movement of inter- regional traffic in countries like India.One of the innovative ground improvement techniques practised all over the world is the use of geosynthetics, which include geotextiles, geomembranes, geogrids, etc . They offer the advantages such as space saving, enviromnental sensitivity, material availability, technical superiority, higher cost savings, less construction time, etc . Because of its fundamental properties, such as tensile strength, filtering and water permeability, a geotextile inserted between the base material and sub grade can function as reinforcement, a filter medium, a separation layer and as a drainage medium. Though polymeric geotextiles are used in abundant quantities, the use of natural geotextiles (like coir, jute, etc.) has yet to get momentum. This is primarily due to the lack of research work on natural geotextilcs for ground improvement, particularly in the areas of un paved roads. Coir geotextiles are best suited for low cost applications because of its availability at low prices compared to its synthetic counterparts. The proper utilisation of coir geotextilcs in various applications demands large quantities of the product, which in turn can create a boom in the coir industry. The present study aims at exploring the possibilities of utilising coir geotextiles for unpaved roads and embankments.The properties of coir geotextiles used have been evaluated. The properties studied include mass per unit area, puncture resistance, tensile strength, secant modulus, etc . The interfacial friction between soils and three types of coir geotextiles used was also evaluated. It was found that though the parameters evaluated for coir geotextiles have low values compared to polymeric geotextiles, the former are sufficient for use in unpaved roads and embankments. The frictional characteristics of coir geotextile - soil interfaces are extremely good and satisfy the condition set by the International Geosynthetic Society for varied applications.The performance of coir geotextiles reinforced subgrade was studied by conducting California Bearing Ratio (CBR) tests. Studies were made with coir geotextiles placed at different levels and also in multiple layers. The results have shown that the coir geotextile enhances the subgrade strength. A regression analysis was perfonned and a mathematical model was developed to predict the CBR of the coir geotextile reinforced subgrade soil as a function of the soil properties, coir geotextile properties, and placement depth of reinforcement.The effects of coir geotextiles on bearing capacity were studied by perfonning plate load tests in a test tan1e This helped to understand the functioning of geotextile as reinforcement in unpaved roads and embankments. The perfonnance of different types of coir geotextiles with respect to the placement depth in dry and saturated conditions was studied. The results revealed that the bearing capacity of coir-reinforced soil is increasing irrespective of the type of coir geotextiles and saturation condition.The rut behaviour of unreinforced and coir reinforced unpaved road sections were compared by conducting model static load tests in a test tank and also under repetitive loads in a wheel track test facility. The results showed that coir geotextiles could fulfill the functions as reinforcement and as a separator, both under static and repetitive loads. The rut depth was very much reduced whik placing coir geotextiles in between sub grade and sub base.In order to study the use of Coir geotextiles in improving the settlement characteristics, two types of prefabricated COlf geotextile vertical drains were developed and their time - settlement behaviour were studied. Three different dispositions were tried. It was found that the coir geotextile drains were very effective in reducing consolidation time due to radial drainage. The circular drains in triangular disposition gave maximum beneficial effect.In long run, the degradation of coir geotextile is expected, which results in a soil - fibre matrix. Hence, studies pertaining to strength and compressibility characteristics of soil - coir fibre composites were conducted. Experiments were done using coir fibres having different aspect ratios and in different proportions. The results revealed that the strength of the soil was increased by 150% to 200% when mixed with 2% of fibre having approximately 12mm length, at all compaction conditions. Also, the coefficient of consolidation increased and compression index decreased with the addition of coir fibre.Typical design charts were prepared for the design of coir geotextile reinforced unpaved roads. Some illustrative examples are also given. The results demonstrated that a considerable saving in subase / base thickness can he achieved with the use of eoir geotextiles, which in turn, would save large quantities of natural aggregates.

Reinforcement Learning Approaches to Power System Scheduling

One major component of power system operation is generation scheduling. The objective of the work is to develop efficient control strategies to the power scheduling problems through Reinforcement Learning approaches. The three important active power scheduling problems are Unit Commitment, Economic Dispatch and Automatic Generation Control. Numerical solution methods proposed for solution of power scheduling are insufficient in handling large and complex systems. Soft Computing methods like Simulated Annealing, Evolutionary Programming etc., are efficient in handling complex cost functions, but find limitation in handling stochastic data existing in a practical system. Also the learning steps are to be repeated for each load demand which increases the computation time.Reinforcement Learning (RL) is a method of learning through interactions with environment. The main advantage of this approach is it does not require a precise mathematical formulation. It can learn either by interacting with the environment or interacting with a simulation model. Several optimization and control problems have been solved through Reinforcement Learning approach. The application of Reinforcement Learning in the field of Power system has been a few. The objective is to introduce and extend Reinforcement Learning approaches for the active power scheduling problems in an implementable manner. The main objectives can be enumerated as:(i) Evolve Reinforcement Learning based solutions to the Unit Commitment Problem.(ii) Find suitable solution strategies through Reinforcement Learning approach for Economic Dispatch. (iii) Extend the Reinforcement Learning solution to Automatic Generation Control with a different perspective. (iv) Check the suitability of the scheduling solutions to one of the existing power systems.First part of the thesis is concerned with the Reinforcement Learning approach to Unit Commitment problem. Unit Commitment Problem is formulated as a multi stage decision process. Q learning solution is developed to obtain the optimwn commitment schedule. Method of state aggregation is used to formulate an efficient solution considering the minimwn up time I down time constraints. The performance of the algorithms are evaluated for different systems and compared with other stochastic methods like Genetic Algorithm.Second stage of the work is concerned with solving Economic Dispatch problem. A simple and straight forward decision making strategy is first proposed in the Learning Automata algorithm. Then to solve the scheduling task of systems with large number of generating units, the problem is formulated as a multi stage decision making task. The solution obtained is extended in order to incorporate the transmission losses in the system. To make the Reinforcement Learning solution more efficient and to handle continuous state space, a fimction approximation strategy is proposed. The performance of the developed algorithms are tested for several standard test cases. Proposed method is compared with other recent methods like Partition Approach Algorithm, Simulated Annealing etc.As the final step of implementing the active power control loops in power system, Automatic Generation Control is also taken into consideration.Reinforcement Learning has already been applied to solve Automatic Generation Control loop. The RL solution is extended to take up the approach of common frequency for all the interconnected areas, more similar to practical systems. Performance of the RL controller is also compared with that of the conventional integral controller.In order to prove the suitability of the proposed methods to practical systems, second plant ofNeyveli Thennal Power Station (NTPS IT) is taken for case study. The perfonnance of the Reinforcement Learning solution is found to be better than the other existing methods, which provide the promising step towards RL based control schemes for practical power industry.Reinforcement Learning is applied to solve the scheduling problems in the power industry and found to give satisfactory perfonnance. Proposed solution provides a scope for getting more profit as the economic schedule is obtained instantaneously. Since Reinforcement Learning method can take the stochastic cost data obtained time to time from a plant, it gives an implementable method. As a further step, with suitable methods to interface with on line data, economic scheduling can be achieved instantaneously in a generation control center. Also power scheduling of systems with different sources such as hydro, thermal etc. can be looked into and Reinforcement Learning solutions can be achieved.