Abstract

The current thesis is aimed at modeling the electrophysiological processes of neurons and myocytes using fractional differential equations (FDE). Using the framework of fractional differential equations, the refractory period, of the action potential generated in a biological cell can be modeled more accurately than that obtained from existing models based on ordinary differential equations (ODE). Electrophysiological processes pertaining to bioelectric phenomenon are of great significance to biomedical engineers because these potentials are recorded in modern clinical practice for medical diagnostic applications on a regular basis. Some of the most common examples of such signals are ECG (Electrocardiogram), EMG (Electromyogram), EEG (Electroencephalogram), ENG (Electroneurogram), EOG (Electro-oculogram), ERG (Electroretinogram). In order to reproduce experimental data of bioelectric phenomena quantitatively, several mathematical models have been proposed. Over the last few decades attempts have been made to make those models more and more accurate. One of the pioneering works in this field was done by Hodgkin and Huxley who developed the model of the axon from the Loligo squid. The Hodgkin-Huxley (H-H) model, although qualitatively correct, the H-H model cannot be applied to model biopotentials in other types of membranes. The cause for this limitation is that the H-H model takes into account only two ions (Na+ and K+), each with a single type of voltage-sensitive channel. However, it has been shown that other types of ions such as Ca2+ may be play a significant role in determining the biopotentials. In addition to the above it has also been empirically established that for every type of ion, there can be a variety of channels. The cardiac action potential, for example, shows how action potentials of various shapes can be generated in membranes with voltage-sensitive Ca2+ channels and different types of Na+/K+ channels. Inspired by the Hodgkin-Huxley model several models of the cardiac myocytes were developed. Though the models were fairly accurate, none of them were able to model the refractory period, duration, of the action potential accurately. The reason for this is that all the differential equation models assumed the membrane capacitance to be ideal, i.e., without any charge leakage, whereas in nature a capacitor can never be ideal. The current-voltage relationship of a capacitor cannot be described by an ordinary differential equation (ODE) accurately. The leaky nature of the capacitor can be modeled more accurately using a fractional differential equation (FDE). The order of the FDE can be varied to obtain accurate values of refractory period, as measured empirically. In the current thesis, we aim to obtain a mathematical relation between refractory period and order of FDE. Using this relation the refractory period corresponding to a given fractional order can be predicted without the need to simulate the mathematical model. Using this relation the refractory period duration for fractional orders above 1 has also been predicted. This was not possible due to the physical limitation that the order of the FDE cannot exceed the number of initial conditions available, and we have only one set of initial conditions to solve the model equations. The Hodgkin-Huxley electrical circuit model of the neuron has also been modified by adding a thermopile, to account for the phenomenon of heat reabsorption in the neuron after an action potential spike has subsided.

Abstract

The modern day societies are becoming more and more dependent on electrical energy. It is the most important part of our day to day life whose demand is increasing continuously. Even with today's advanced power system technologies the losses in the system are very high which are needed to be reduced for higher efficiencies and better receiving end voltages. This is experienced even more in distribution system where there is a direct impact on customer, owing to long radial lines (unique path from any given load bus to source). Hence, a solution that addresses both the concerns i.e. reduced distribution losses and better voltage profile in the system is of great interest. This can be achieved by harnessing renewable resources and integrating them into the system as distributed generation. The size/installed capacity of renewable generation is also an important factor that cannot be neglected due to the intermittent nature of availability of many renewable sources like solar and wind. Renewable energy sources can be deployed and used in the distribution systems with more eagerness in present day scenario. They impose a considerable impact on the system and customer owing to the intermittence nature. Thus, the problem of incorporating renewable generation into distribution system becomes more complex. This thesis proposes a methodology for integrating the renewable generation at optimal location with reduced distribution losses. The total capacity of renewable generation to be installed is calculated for the given distribution network. The generation capacity is decided so as to maintain all the bus voltages in acceptable limits based on 24 hour system load. The approach involves a process of finding bus voltages, line currents and power loss in the system by distribution load flow. The proposed work uses radial distribution load flow algorithm for the chosen distribution test networks. This thesis considers integration of renewable energy sources namely Wind, Solar photovoltaic and Biomass as distributed generation. Each type of renewable energy source is mathematically modeled and the exact amount of power output is calculated for the given meteorological conditions. The proposed methodology is validated by implementing on standard radial 33 and 69 bus test systems. The simulation results are presented for both with and without inclusion of renewable energies and discussed.

Abstract

Glaucoma is an eye disorder which is prevalent in the aging population and causes irreversible loss of vision. Hence, computer aided solutions are of interest for screening purposes. Glaucoma is indicated by structural changes in the optic disc (OD), loss of nerve ?bres and atrophy of the peripapillary region of the OD in retina. In retinal images, most of these appear in the form of subtle variation in appearance. Hence, automated assessment of glaucoma from colour fundus images is a challenging problem. Prevalent approaches aim at detecting the primary indicator, namely, the optic cup deformation relative to the disc and use the ratio of the two diameters in the vertical direction, to classify images as normal or glaucomatous. We explore the use of global motion pattern-based features to detect glaucoma from images and propose an image representation that serves to accentuate subtle indicators of the disease. These global image features are then used to identify normal cases effectively. The proposed method is demonstrated on a large image dataset consisting of 1845 images annotated by 3 medical experts. The global approach is extended to detect atrophy and two hierarchical system designs are proposed. In the ?rst design, only global analysis is used, while in the second both global and local analysis are employed. In the ?rst design, the ?rst stage is based on features capturing information mainly of primary indicators while the second stage is based on features extracted for detecting atrophy (secondary visual indicator). The second design attempts to combine the strengths of global and local analysis of the OD region. Global features are used to remove as many normal cases as possible in the ?rst stage and local features are used to perform a ?ner classi?cation in the second stage. This system has been tested on 1040 images with ground truth collected from 3 glaucoma experts. The results show the hybrid approach offers a good solution for glaucoma screening from retinal images (more...)

Abstract

Ray Tracing is one of the most important rendering techniques used in computer graphics. Ray traced images are more accurate and photo-realistic as compared to direct rendering. Ray Tracing was earlier considered impractical for rendering scenes at interactive rates because of its high computational cost. However, with the advancements in modern Graphics Processing Units (GPU) and CPUs, ray tracing at interactive rates has now become possible. Parametric patches have been widely used in many fields to describe a model accurately. They provide a compact and effective way of representing an object and also possess the ability to remain curved on zooming. Ray Tracing of parametric surfaces was considered to be a static process because of the high complexity of intersection algorithms. With advancements in ray tracing techniques and high compute power devices, recent works on ray tracing parametric surfaces have reported near interactive results. We present a scheme for interactive ray tracing of Bezier bicubic patches using Newton iteration in this dissertation. We use a mixed hierarchy representation as the acceleration structure. This has a bounding volume hierarchy above the patches and a fixed depth subpatch tree below it. This helps reduce the number of ray-patch intersections that needs to be evaluated and provides good initialization for the iterative step, keeping the memory requirements low. We use Newton iteration on the generated list of ray patch intersections in parallel. Our method can exploit the cores of the CPU and the GPU with OpenMP on the CPU and CUDA on the GPU by sharing work between them according to their relative speeds. A data parallel framework is used throughout starting with a list of rays, which is transformed to a list of ray-patch intersections by traversal and then to intersections and a list of secondary rays by root finding. We are able to significantly outperform multi-core CPU implementation and previous GPU implementation using the mixed hierarchy model. Shadow and reflection rays can be handled exactly in the same manner as a result. The secondary ray list is again sent to the starting of the algorithm to perform mixed hierarchy traversal and intersection tests. We perform fixed depth multiple bounce ray tracing. We also show how our method extends easily to generate soft shadows using area light sources. These effects provide higher realism to the ray traced images. We render a million pixel image of the Teapot model at 125 fps on a system with an Intel i7 920 and a Nvidia GTX580 for primary rays only and at about 65 fps with one pass of shadow and refection rays. We are able to ray trace bigguy in a box scene with multi-bounce at near interactive rates. We get a speed up of about 5-30x for our hybrid Newton's method implementation over our optimized CPU implementation and about 20-50x over previous GPU implementation of Kajiya's method to ray trace Bezier surfaces. Traversing the mixed hierarchy is the most time consuming step of the algorithm. We expect to see better performance with greater cache size. The hybrid model would be optimal for systems with equal compute power of CPU and GPU. The proposed model is suitable for parallel architecture, hybrid systems and multi-GPU systems. Global illumination effects have recently started gaining popularity with the progress in parallel architecture. We extend our algorithm to global illumination effects to demonstrate its capabilities. Global illumination effects have not been reported for parametric surfaces. We perform path tracing by tracing a large number of rays per pixel for a fixed depth. Number of samples greatly increase the quality of the image generated. We also perform more advanced effects like ambient occlusion, depth of field, motion blur and glossy surface. We are able to path trace a $512 imes 512$ image with 1000 samples per pixel in about 165 seconds. We report timings for other advanced effects. We find that ray coherence is essential for optimal performance when ray tracing Bezier surfaces on the GPU. Size of the dataset also plays a small part in the overall rendering times. The work done in this dissertation should serve as the starting point to optimally render Bezier surfaces with advanced global illumination techniques.

Abstract

Computational advertising, popularly known as online advertising or web advertising, refers to nding the most relevant ads matching a particular context on the web. It is a scienti c sub-discipline at the intersection of information retrieval, statistical modeling, machine learning, optimization, large scale search and text analysis. The core problem addressed in Computational Advertising (CA) is of match making between the ads, context and user. Depending upon the context, ads are shown based on two kinds of targeting { content targeting and user targeting. Content targeting involves displaying ads relevant to the content being targeted. Retrieving ads for a query on a search page (known as Sponsored Search) or for a third party web-page content (known as Contextual Advertising) are instances of content level targeting. User level targeting involves modeling user's recent interests based on their activities and showing ads relevant to user's interest. Recent outburst in the usage of social networks and recommender systems has made targeting of ads based on user's prole a popular method. Within user modeling, ads can usually be targeted based on user's recent activity or based on user's social contacts. Net ix, Amazon, StumbleUpon are some examples that recommend ads/sponsored content based on user's recent activities. Facebook is a common example of targeting ads based on user's social contacts and reach in a social network. This thesis aims at analyzing the problems and contributing towards solutions to these two types of targeting. Specically, the thesis work plans on leveraging various user interactions, advertiser related and contextual information for dierent types of targeted advertising. The research work deals with identication of various factors that contribute in retrieval and ranking the most relevant set of ads that match best with the context. For this purpose, information associated with the user, publisher and advertiser is leveraged. Given the proprietary nature of datasets on ad targeting, there is a scarcity of public datasets. Experiments involved in this thesis provide substantial amount of labeled data that can also be used to tune future ad targeting systems. The research work in this thesis primarily focuses on the problem of retrieving relevant ads for dierent content types. This thesis proposes a couple of semantic solutions to the problem of retrieving ads for a given content. As ads that are being retrieved are short, often there is a vocabulary mismatch between the query and the ads. This problem renders traditional information retrieval techniques insucient for the task of ad retrieval. We propose two techniques based on modeling the latent topics/concepts from the ads and the content for retrieving relevant ads. The idea is to transform the ads and the content into lower topical/conceptual dimension and retrieve ads in that lower dimensional space. This lower dimensional transformation helps reduce vocabulary mismatch to some extent. The thesis begins with simplistic topic modeling approach and then builds on top of it with the conceptual modeling approach. These techniques are studied extensively for a variety of dierent content types like user search query, third party web pages and microblogs such as tweets. The proposed semantic techniques are compared with various state-of-the-art retrieval techniques for the problem of ad retrieval. The proposed techniques perform better than all the baseline state-of-the-art approaches. The performance improvement for the proposed best performing semantic method over the best performing baseline ranges from around 10% to as much as 28%. This performance improvement of the proposed techniques is consistent over dierent content types like search query, web pages and microblog. Considering the hardness of the ad retrieval problem, the percentage improvement shown by the proposed method over dierent content types is impressive. In addition, we also study the combination of various techniques on this problem of retrieving ads. It turns out that various combinations of semantic techniques proposed in this thesis also perform better than combinations of dierent baseline approaches. This performance improvement was consistently observed for several metrics like precision @ n, precision-recall etc. All the performance improvements were found to be statistically signicant at very good signicance levels (96% to 99%). The thesis also studies matching of ads from user targeting perspective. We look at the targeting of users based on their actions and their prole. The idea here is to model the user in order to facilitate better matching of relevant ads with the user context. We study following dierent user contexts { user and their prole, user's social contacts, user's action. The user targeting study aims at exploring various factors pertaining to a user that aect the targeting of ads. To the best of our knowledge,

Abstract

Technology has enabled the rapid creation and easy storage of vast amounts of digital spoken content. In this context, a reliable and fast way to search through spoken content has become a hot topic of research in speech technology. Query-by-Example Spoken Term Detection (QbE-STD) is a framework for searching a speech database using spoken queries. The aim of this work is to develop an improved dynamic time warping (DTW) based QbE-STD by exploiting syllable boundary information. Within this scope, novel contributions are made within two research themes, i.e., locating the syllable nuclei and thus subsequently locating the syllable boundaries and, using these syllable boundaries to improve the accuracy of the DTW based QbE-STD system. Speech can be segmented into syllables by identifying the syllable nuclei, which are points of high sonority. The excitation peaks in the linear prediction (LP) residual and the formant peaks can be interpreted as perceptually significant point features which contribute to the loudness of speech. In this paper, the use of these two point features is described for the use of detecting syllable nuclei. Each of these evidences contain information about different aspects of speech production, namely the glottal vibrations and the time varying vocal tract system. Thus it is possible that they contain complementary information about the syllable nuclei. Performance of the proposed syllable nuclei detection algorithm is evaluated for the TIMIT, Switchboard and the NTIMIT corpus. The proposed method performs comparably against two other state of the art syllable nuclei detection methods, and it is very fast requiring no training. Important acoustic events in speech occur at the transition boundaries between two sounds. These acoustic events carry sufficient information so as to be useful for QbE-STD. The syllable boundaries and the transition boundaries within the syllable are anchor points where major acoustic events occur. In this work, the use of these two types of anchor points are explored for feature selection. This initial feature selection method serves as a way to prune the search space for QbE-STD for later re-ranking. Results are demonstrated for the MediaEval2012 Spoken Web Search database. The performance after re-ranking the pruned results are on par or better than the standard results returned by the DTW technique.

Abstract

In recent years, the microblogging services like Twitter has become a major tool for sharing events, expressing opinions and communicating with friends. Several thousand of microblogs are posted each second describing the ongoing events around the world. Because of this recent rise in the popularity and size of social media, there is a growing need for system that can extract useful information from social media. The use of limited words (140 characters), special vocabulary with incomplete sentences and emoticons, heavy use of abbreviations and misspellings make traditional text analysis methods less effective on Twitter. Entity Linking is the task of extracting semantics from text by linking the entities present within the text to the knowledge base. For example, given a text “India is my country”, we need to extract important mentions like ‘India’ and link them to their corresponding Wikipedia pages. Entity Linking is an important task as it not only provides important meta-information about the linked entities but also helps us in extracting the relations among various entities present within the text. Meta-information and relations are really helpful in extracting other important information like documents’ topics etc. Since this also removes the problem of synonymy and polysemy, therefore it is more easy for computers to analyse the text. So, entity linking is really helpful in making computers understand the text usually meant for human consumption. In this thesis, we device a method for efficient entity linking. We have devised various features that helps in detecting and linking entities present within the tweet. Firstly, the mentions present within the tweets are identified. We experimented various approaches like heuristics based approach, probabilistic approach and sequence labelling based approach for detecting mentions in the tweet. Finally, we proposed a classification approach for mention detection. Along with identification, the detected mentions are also ranked based on their predicted informativeness content. The user can select the mentions by setting the threshold on the ranks based on application. All the approaches are evaluated using various datasets like NEEL Challenge Dataset, User Dataset, TagMe Dataset and IITB dataset. Classification based methods performs best on all the datasets with performance of 0.59 F-measure on NEEL Dataset. The detected mentions are then disambiguated on the basis of context. Disambiguation means selecting the best entity from the set of candidate entities for the given mention. Due to limited characters, context is usually not sufficient to disambiguate all the entities present within the tweet. So, we have also explored users’ interests and recent tweet information to improve the disambiguation on the short and noisy tweets. The information provided by users’ previous tweets is modeled to build users’ topic of interests. These interests along with the context is then used to disambiguate the entities in the users’ new tweet. We also explored trending topics information (at a particular time on Twitter) to improve entity disambiguation in the tweets. The user based approach is tested using user dataset collected from around 200 tweets of 20 users with two context modeling systems. Recent tweet information based approach is tested using the standard NEEL Dataset and tweets from recent events collected in June 2014. Experiments show that extra information based on users’ previous history and recent tweets helps in improving the performance of entity disambiguation systems. Hashtags are semantico-syntactic constructs used across various social networking and microblogging platforms to enable users to start a topic specific discussion or classify a post into a desired category. Segmenting and linking the entities present within the hashtags could therefore help in better understanding and extraction of information shared across the social media. However, due to lack in space delimiter, extracting semantics from hashtags is a non-trivial task. Most of the current state-ofthe- art social media analytics systems like Sentiment Analysis and Entity Linking tend to either ignore hashtags, or treat them as a single word. We present a context aware approach to segment and link entities in the hashtags to a knowledge base (KB) entry, based on the context within the tweet. Our approach segments and links the entities in hashtags, such that the coherence between hashtag semantics and the tweet is maximized. As the segmentation and linking is jointly modeled, the system chooses the segment that maximizes the relatedness between tweet and the hashtag. So, the best segmentation is picked based on the context, if multiple segments are possible. The approach is tested on two different datasets, one is synthetically created using NEEL Dataset and the other is manually created. We have achieved the P@1 score of 0.914 on NEEL Dataset and

Abstract

Statistical machine translation (SMT) depends heavily on parallel training data available for a language pair. This thesis addresses problems of data-sparsity and translation quality of a SMT system for English-Hindi and Indo-Aryan language pairs. We also show use of machine translation (MT) in improving speech recognition output in the scenario of a computer assisted translation (CAT) system. SMT depends heavily on parallel training data available for a language pair. We describe a MT system for English to Hindi in which we use a pre-processing module to transform English data into Hindi-Order to get better phrase alignments. Morphologically rich languages require large amounts of parallel data to adequately estimate parameters in a statistical MT system. We explore alternate strategies, given that the collection of large parallel corpora is expensive. Firstly, we explore the use of distributed representations in a language model in a Recurrent neural network based Language Model (RNNLM) framework with different linguistic features and secondly, we explore the use of lexical resources such as WordNet to overcome sparsity. The techniques used above rely on large mono-lingual data for the target language and also requires WordNet of the source language. WordNet is not available for Indo-Aryan languages except Hindi and there are not many sources available for extracting clean mono-lingual data. So we improved the translation quality of Indo-Aryan MT systems by exploring popular available techniques called ``Triangulation'' and ``System Combination'' to reduce problem of data sparsity and improve quality of the system. MT systems for other Indo-Aryan language pairs which are trained on small parallel corpora are improved using various recently developed techniques in the field of SMTs. Triangulation is a technique which has been found to be very useful in improving the translations when multilingual parallel corpora are present. Triangulation is the process of using an intermediate language as a pivot to translate a source language to a target language. We have used phrase table triangulation instead of sentence based triangulation as it gives better translations. As triangulation technique explores additional multi parallel data, it provides us with separately estimated phrase-tables which could be further smoothed using smoothing methods. The subsequent approach explores the various system combination techniques through which these triangulated systems can be combined to improve the translations. In concluding chapter of the thesis, we show the application of MT in improving the Automatic Speech Recognition (ASR) output in the scenario of ocite{*}a CAT system where a human translator translates a source language string into a target language string using different input methods such as speech and typing. In this work, we have improved the performance of speech recognition of a translator speaking in the target language, taking the advantage of source language string and synset information from WordNet. We used MT system to translate the source language string to target language. We use the semantic information we get for the words in the translated string from WordNet to bias the speech recognizer towards the gained knowledge. We perform different experiments including variation of number of hypothesis of MT and also different techniques of incorporating the semantic information. Overall we outperformed the baseline system having no semantic information by the increase in word accuracy of 1.6% for the Hindi ASR in English-Hindi system.

Abstract

Binding of gold clusters (Au3 and Au4 ) with DNA bases and extended DNA bases (x-bases) has been studied in this work. An attempt to locate the energetically low-lying minimum structures in case of Aun clusters (n=3-8) was carried out using density functional theory. Both the studies have been performed by employing high quality basis set. A number of geometrical isomers, in case of Aun clusters, was studied. The relative stabilities were then compared to determine the minima. Electronic properties such as HOMO-LUMO gaps, polarizability and dipole moments,etc., were also investigated. In gold-nucleobase complexes, all possible orientations of Au clusters with respect to the titratable sites of bases were examined. The geometric and electronic features of these complexes provided evidence for existence of non-conventional hydrogen bonds. This was further substantiated via normal mode and NBO analysis. The nucleobases, both canonical and extended forms, form stable complexes with the gold clusters considered here. Upon complexation, a marked decrease in the HOMO-LUMO gaps and ionization potentials was observed, which was more profound in case of x-bases. This study attempts to provide an insight into the stability of different-sized gold clusters and demonstrates that combining structural modifications to DNA bases and subsequent binding to gold nano particles provides an opportunity to modulate and achieve desired optico-electronic properties.

Abstract

Arithmetic and Logic Unit (ALU) is a critical component of any CPU. In ALU, adders play a major role not only in addition but also in performing many other basic arithmetic operations like subtraction, multiplication, etc. Thus realizing an efficient adder is required for better performance of an ALU and therefore the processor. Research started in late 1950s on designing efficient adder algorithms and their hardware implementation. Many designs based on serial and parallel structures have been proposed to optimize different parameters from time to time. The first contribution of this thesis is the development of an efficient adder architecture that addresses the problems for higher bit operand lengths like fan-out, wiring complexity, etc. Another important element in an ALU after adder is a multiplier. In multipliers, for reducing partial products and computing final result, multi-operand adders and fast adders are required. A special structure known as counters/compressors are typically used for designing multi-operand adders. Counters are multi-input, multi-output combinational logic circuits, which determine the number of logic 1’s in their input vectors, and generate a binary coded output vector that corresponds to this number. Large parallel counters like (15, 4), (32, 5), etc. can be constructed using this small counter and similar approach can be adopted in the case of compressors. The second contribution of the thesis is development of efficient counters and compressors for better performance of multiplier. Apart from adders and multipliers in arithmetic units, elements like, incrementer/decrementer (INC/DEC) also play a major role in an ALU and also in address generation unit. A loop algorithm, for example, often needs a increment/decrement. These operations can be realized using adders but with a cost in terms of power and area. Therefore, standalone designs or unified designs for INC/DEC are required for low power applications. The third contribution of this thesis is the design of a multi-functional INC/DEC/2's complement/Priority encoder circuit. A design for binary INC/DECs is presented that is efficient in terms of speed without compromising on power. The need to have hardware support for decimal arithmetic is increasing in recent years because of the growth in decimal data processing in commercial, financial and internet-based applications. To facilitate binary computations on the same hardware, a reconfigurable approach needs to be adopted. The fourth contribution of this thesis is the design of a new architecture for efficient Binary Coded Decimal (BCD) addition/subtraction that can be configured to perform binary addition/subtraction also. The architecture has been designed keeping in view the signed magnitude format where the adder logic itself detects the larger operand and carries out corresponding operations. Finally, novel versions of two widely used arithmetic blocks i.e., multiplier and floating point adder, are designed. Efficient and proven basic functional units described above are used to implement these blocks. Simulations of these blocks have been carried out and comparisons made with existing designs that clearly demonstrate the efficiency of proposed units. Finally, a segment of a core of a processor is designed with incorporating all the above elements resulting in an efficient architecture.