Abstract

Representing knowledge in a diagrammatic form has been a long-standing goal of humanity. Early efforts in the field of knowledge representation, such as symbolic logic and semantic net, have led to the development of The Semantic Web, which provided a strong foundation for the development of Knowledge Graphs (KGs). KGs were first introduced in 1986 and were popularized recently in 2012 with the introduction of the Google knowledge graph. Most KGs suffer from the problem of incompleteness, which has led to many efforts in the field of KG completion. In this thesis, we address a specific subproblem from this field called link forecasting. Link forecasting is the problem of predicting future links in a given temporal knowledge graph (TKG) using the existing data. Although several link forecasting frameworks exist in the literature, most previous studies suffer from reduced performance as they cannot efficiently capture the time dynamics of a TKG. We present a novel rule-based link forecasting framework by introducing two new concepts: relaxed temporal random walks and link-star rules. The former concept involves generating rules by performing random walks on a TKG, considering the real-world phenomenon that the order of any two events may be ignored if their occurrence time gap is within a threshold value. The latter concept defines a class of acyclic rules generated based on the natural phenomenon that history repeats after a particular time. Our framework also accounts for the problem of combinatorial rule explosion, making our framework practicable. Experimental results demonstrate that our framework outperforms the state-of-the-art by a substantial margin

Abstract

Hedgehog (HH) signaling pathway is the cellular pathway for patterning and morphogenesis. It is controlled by two membrane proteins, the HH receptor Patched1 (PTCH1), a 12-pass transmembrane (TM) protein, and Smoothened (SMO), a 7-pass transmembrane (TM) signal transducer. SMO transduces the signal from the extracellular (EC) region to the cytoplasmic (CP) region. Binding of HH ligand to PTCH1 receptor allows SMO entry to the primary cilium (PC) and activates the downstream signaling process. The malfunctioning of SMO results in misregulation of the HH signaling pathway, which contributes to birth defects and various cancers. The story has evolved from initial conjectures involving direct communication of PTCH1 with SMO to indirect communication between the two TM proteins. In the process, some of the old theories have been discarded in recent literature. However, the underlying mechanism of how PTCH1 controls SMO is still not very clear, and thus constitutes an important area of research. Based on available evidence, the consensus appears to be limited to the assertion that, driven by HH signaling, PTCH1 regulates the activity of SMO activity through the intermediacy of some small molecules (such as cholesterol). In addition, it appears that the lipid bilayer composition affects SMO orientation and function. In this context, several earlier hypotheses involving the primary role of cholesterol are being disputed. At the same time, several HH pathway inhibitors viz., cyclopamine, taladegib, and vismodegib, which target SMO have been reported. Notwithstanding their potential utility in the treatment of cancer, long-term administration of these drugs was reported to lead to the development of resistance. The broad aim of my research is to build upon the existing molecular-level understanding of the functioning of the HH pathway, with a view to facilitate the formulation of newer approaches towards therapeutic intervention. The focus of my research is to investigate the structure and domain organization of SMO, which is a class F GPCR. This thesis attempts to provide some insights into the functioning of the SMO receptor, using molecular dynamics-based approaches. It is shown that SMO is activated by cholesterol molecules and that its dynamics is influenced by other membrane lipids such as PI4P. The specific objectives include 1. To identify important motifs, especially including, but not limited to, the lipid-binding sites/motifs. 2. To study the conformational dynamics of SMO, to help understand its molecular behavior and also the role of the composition of the primary cilia. 3. To study the regulation mechanisms in SMO by exploring the different cholesterol-bound structures of SMO.

Abstract

Data mining is a collection of algorithms for extracting valuable insights from large amounts of data. Data mining based approaches are being employed to improve the performance of decision support systems in the fields of customer relationship management, inventory management, fraud detection, surveillance, and recommendation systems. Pattern mining is an important task of data mining, which involves identifying significant associations in transactional databases. These associations reveal valuable trends and relationships that aid businesses in improving efficiency. The field of pattern mining has started with the model to extract frequent patterns from large transactional data. The model of frequent patterns has become popular and resulted in the development of algorithms to improve the performance of several applications, like recommendation systems. In the literature, encouraged by the potential of pattern mining, active research is going on to investigate new pattern mining models such as periodic, utility, coverage, and correlated patterns. In this thesis, we propose an improved approach to mine periodic patterns from temporal databases. The model of periodic patterns facilitates the discovery of recurring behaviours and trends in the temporal databases. The model of periodic pattern, which we call partial periodic pattern (3P), captures periodic associations subject to periodic support (P S) and inter-arrival time (IAT) constraints. Here, the IAT value is the time difference between the successive occurrence of a pattern. The percentage of occurrences of a pattern in a database that satisfies the user-given maximum IAT constraint is called P S. Overall, all patterns that satisfy user given maximum IAT (maxIAT) and minimum PS (minP S) are called 3P s. In this thesis, we address the following issue of the 3P s model. It was observed that if we set the minP S value too low, the number of 3Ps explodes. On the other hand, if set minP S value high, several interesting 3Ps will be missed. We call this problem as a rare item problem. To address this issue, in this thesis, we have proposed an improved model and present an improved depth-first search algorithm to extract 3Ps. In the proposed model, we introduce a new measure called periodic-conf idence, which satisfies both null-invariant and anti-monotonic properties. We also propose a better depth-first search algorithm to mine 3Ps. The proposed algorithm uses irregularity pruning to reduce the search space and computational cost while maintaining the same amount of information. Through experimental results, we show that the proposed algorithm is efficient and scalable. We illustrate the usefulness of the discovered patterns through case studies on air pollution and traffic congestion databases. Extracting interesting trends from large temporal databases in different domains is an active research area. We hope that the proposed approach could facilitate the development of improved approaches to extract regular trends for fraud detection from sensor (surveillance) databases and loyalty mining in e-commerce systems.

Abstract

The current research aims to develop a multichannel phonocardiography system to improve the efficiency in detecting low-frequency cardiac auscultation and diagnosis of heart failures. In particular, the research work has been carried out with respect to two aspects. Firstly, a hardware prototype of a MEMS-based multichannel phonocardiography system has been developed, along with the placement of microphones and methods for heart sound localization to enhance signal quality. Secondly, proposed the signal processing algorithm for heart sound segmentation to identify the low-frequency components and neural network-based data analytic technique for feature extraction and implemented the proposed algorithms on SoC-FPGA to differentiate the normal and abnormal heart sound components. Cardiac auscultation is one of the non-invasive methods to diagnose heart abnormalities. Along with four significant heart sounds (S1, S2, S3, and S4), other murmur sounds are generated due to pathological conditions. These anomalies help in proper diagnosis and prevent the possibility of heart failure. MEMS are becoming the most popular due to their size and free ambient noises and are vital in developing noninvasive diagnostic instruments. In addition, The development of cardiac auscultation devices advantages significantly from using MEMS microphones. This research has developed the MEMS-based phonocardiography instrument to capture and analyze the heart sound S3, S4, and murmur components at cardiac auscultation points. The main contribution of this thesis is developing the mathematical model for source localization algorithms to derive microphone positions with a high signal-to-noise ratio (SNR). The proposed cross-correlation method improves the system's sensitivity in detecting low-frequency cardiac sound components (S1, S2, S3, S4 and stenosis, regurgitation murmurs). The segmentation approach combining wavelet and Shannon energy was implemented on Field Programmable Gate Arrays (FPGAs) to categorize the heart sound components. The proposed segmentation algorithm revealed an excellent sensitivity of 99.17 percent and a detection error rate of 1.5 percent. Accurate measurements of the cardiac components were obtained using a combination of traditional statistical approaches and neural network based algorithms from multiple signals received simultaneously from several PCG systems. The proposed multichannel phonocardiography system analysis the cardiac sound components using artificial neural networks (ANN). The Inverse delayed (ID) function model of a neuron is used to compute synaptic weights after being simulated in MATLAB. The proposed ANN model was implemented in a Field Programmable Gate Array (FPGA). An SoC FPGA (ZYNQ SoC) performs most of the required data processing, eliminating the need for robust and expensive computer systems. Using regression analysis, a statistical model was developed to identify abnormal heart sounds from the captured signals. The receiver operating characteristic (ROC) curve has been used to evaluate the performances of the proposed model. The ANN system examined both abnormal and normal samples, and experimental results revealed a good sensitivity of 99.1% and an accuracy of 0.9. The research concluded that signal acquisition from multiple sensors and source localization methods produces a high-quality signal suitable for analyzing low-frequency cardiac sounds (S1, S2, S3, S4, stenosis, regurgitation murmurs). In addition, the proposed ANN classification method, based on the inverse delayed (ID) function model of neuron, can resolve combinatorial optimization problems. The negative resistance of the ID model can destabilize a neural network's stable equilibrium points, reducing the possibility of unknown values in suboptimal synaptic weight solutions obtained using an ANN based on a traditional neuron model. Furthermore, the repeatability and reproducibility measurements are used for the performance analysis.

Abstract

In this thesis, we consider three problems relating to coding theory. The first problem is the application of codes to cryptography, while the other two problems tackle the ongoing developments of coding theory. We consider the secret sharing problem in the context of distributed storage. In this setup, there is a dealer, n storage nodes, m secrets, and m t
users. The dealer wants to share with each user a t-subset of secrets via storage nodes while ensuring weak secrecy condition. The user downloads shares from the storage nodes based on the designed storage structure and reconstructs its secrets. We extend upon a recent framework [1] (considering t = 1) to the case where each user requests a subset of secrets. We identify a necessary condition on the storage structure to ensure weak secrecy. A distributed secret sharing protocol (DSSP) encodes secrets into shares and distributes them onto the storage node. The number of shares created per secret is defined as the storage overhead of the DSSP, which is at least one to ensure the correctness condition. The goal is to design a DSSP with optimal storage overhead. We relate the storage structure of a DSSP to t-disjunct matrices used in group testing. Using a storage structure obtained from disjunct matrices, we design a weakly secure DSSP that achieves optimal storage overhead. Using several constructions for t-disjunct matrices, we compare their performance in terms of the number of storage nodes and communication complexity (total number of shares needed to be downloaded from storage nodes). The rate of a secret is the size of the secret normalized by the size of a share, and the capacity region of the problem is the set of all achievable rate tuples. For a specified storage structure, we characterize the capacity region of the problem. Next, we consider the unique decoding algorithms for Reed-Muller (RM) codes. RM codes were shown to be capacity-achieving on the binary memoryless symmetric channels using bitwise maximuma-posteriori (bit MAP) decoding [2]. We propose an RXA-Syndrome decoding algorithm for RM(m, m− 4) code that exploits the large symmetry in the automorphism group of RM codes and the optimality of the maximum likelihood (ML) decoder. Codes that meet the Singleton bound with equality are called Maximum Distance Separable (MDS) codes. In a sense, MDS codes achieve the optimal trade-off between the size of the code and its minimum distance. However, unique decoding of MDS codes can correct errors only up to a certain limit which is related to the minimum distance of the code. In the third problem, we consider the construction of codes that can correct errors beyond this limit. To this end, list decoding was introduced recently, which returns a list of codewords containing the correct codeword. We derive necessary and sufficient conditions for a code to be (t, L)-list decodable. Recently, list decodable codes that meet generalized Singleton bound were considered as the higher-order generalization of MDS codes. It is important to understand the structural properties of these codes to develop explicit constructions and efficient decoding algorithms. Motivated by these developments, we present new results on the structure of higher order MDS codes. We identify conditions under which (n1, k1)-MDS(ℓ1) codes can be obtained from (n2, k2)-MDS(ℓ2) codes. We also simplify the conditions under which an (n, k)-MDS(k − 1) code is also an MDS(k) code.

Abstract

Invertible convolutions have been an essential element for building expressive normalizing flowbased generative models since their introduction in Glow. Several attempts have been made to design invertible k × k convolutions that are efficient in training and sampling passes. Though these attempts have improved the expressivity and sampling efficiency, they severely lagged behind Glow which used only 1 × 1 convolutions in terms of sampling time. Also, many of the approaches mask a large number of parameters of the underlying convolution, resulting in lower expressivity on a fixed run-time budget. We propose a k × k convolutional layer and Deep Normalizing Flow architecture which i.) has a fast parallel inversion algorithm with running time O(nk2 ) (n is height and width of the input image and k is kernel size), ii.) masks the minimal amount of learnable parameters in a layer. iii.) gives better forward pass and sampling times comparable to other k × k convolution-based models on real-world benchmarks. We provide an implementation of the proposed parallel algorithm for sampling using our invertible convolutions on GPUs. Benchmarks on CIFAR-10, ImageNet, and CelebA datasets show comparable performance to previous works regarding bits per dimension while significantly improving the sampling time.

Abstract

Cocaine use disorder (CUD) is a compulsive urge to seek and consume cocaine despite the inimical consequences. MRI studies from different modalities have shown that CUD patients exhibit structural and/or functional connectivity pathology among several brain regions. Nevertheless, both connectivities are commonly studied and analyzed separately, which may potentially obscure its relationship between them, and with the clinical pathology. Here, we compare and contrast structural and functional brain networks in CUD patients and healthy controls (HC). The existing body of research has predominantly concentrated on studying substance abuse within the Caucasian population. Gaining a comprehensive understanding of the distinctive challenges and cultural environments in which substance abuse disorders emerge among various populations is essential. For this study, we recruited 105 individuals of Mexican descent and carefully examined the T1-weighted, diffusion weighted imaging, resting-state fc-fMRI sequences and clinical metrics of these participants. Based on the previous work [53], we identified regions of interest (ROIs) that are known to have altered structural connectivity in CUD patients and examined their pairwise functional connectivity. The results demonstrated that CUD exhibited stronger connection between the right Posterior Cingulate and right Postcentral than HC which could suggest an increase in interoception potentially associated with compulsion behavior. We computed a battery of graph-based measures from multi-shell diffusion-weighted imaging (tc-dMRI) and resting state fc-fMRI to quantify local and global connectivity. We investigated the differences in functional and structural modalities between the two groups independently. Unimodal analysis showed an increase in the participation coefficient in Striatum among CUD compared to HC in the fc-fMRI modality which has been previously linked to the compulsive drug-seeking behavior observed in CUD. Multimodal fusion is a data driven approach that involves fusion of different brain modalities to gain a more comprehensive understanding of the brain’s structure and function. Multimodal fusion has played an important role in investigating joint functional and structural changes in neurological disorders like Alzheimer’s disease, schizophrenia, and epilepsy. This is the first study that uses multimodal fusion of graph theoretical measures to investigate topological alterations in CUD patients. We specifically used multimodal canonical component analysis plus joint independent component analysis (mCCA+jICA) to evaluate group differences and their association with clinical scores. When performing multimodal fusion analysis, we observed a higher betweenness centrality and lower participation coefficient in CUD patients indicating reorganization of functional networks. In addition to altered straiatal connectivity revealed by the unimodal approach, multimodal fusion revealed latent information about brain regions involved in impairment due to cocaine abuse. Overall, results revealed by our study not only concord with previous research in the field but also uncovered findings which could help in understanding the pathology of CUD and develop better pre-treatment/post-treatment intervention design.

Abstract

Unmanned Aerial Vehicles (UAVs) continue to penetrate diverse sectors, including agriculture, disaster management, communication, transportation, and defense. The robustness and reliability of their operation have become paramount due to the increasing ubiquity of these systems across numerous sectors. UAVs can undergo faults in their sensors, actuators (motors), and even structure - all of these have different levels of severity and effects on the system. Motor faults can be of three main types, namely, Motor Locking, Loss of Efficiency (LoE), and Loss of Actuation (LoA). Motor locking is a case where the actuator keeps spinning at a given RPM and cannot be changed, whereas LoE is a case where the actuator degrades over time and is not able to generate enough torque due to lower RPM. In this thesis, however, we focus on LoA, where an actuator stops working completely (gets stuck at zero RPM) and can result in catastrophic failure in the UAV. This work delves into the dynamics of UAVs under actuator fault conditions, examines the resulting instability issues, and explores potential methods for identification and control reconfiguration for safe operation. Once an actuator failure occurs, the Fault Detection & Isolation (FDI) module should be able to locate the fault and pinpoint it with high accuracy and in a very short time, to allow the UAV to stabilize the fault. Various approaches can be used for the FDI system, namely, rule-based, model-based, and data-driven. Implementation of these methods can be relying on RPM sensors, battery monitors, and software-only solutions. We propose a data-driven software module for this purpose, since it is UAV frame agnostic, does not require any additional hardware (software-only), and can be run with minimal setup. The proposed Rotation Forest based classifier can detect, classify, and report the fault within 120-250 milliseconds of occurrence of the fault. This is an acceptable delay, since we have observed that the vehicle cannot handle delays upwards of 500 milliseconds in simulations. Once a fault is reported by the FDI module, the Fault Tolerant Control (FTC) module reconfigures the control system to stabilize the vehicle and continue the mission, or prevent a crash by peforming a safe landing. This thesis focuses on cases of complete actuator failure in Hexacopter UAVs, specifically, for single motor failure scenarios. The proposed FTC module performs UAV stabilization using control reconfiguration, after fault occurrence. We perform a thorough analysis for the FDI and FTC modules separately, extensively in simulation, and demonstrate the same in real flight tests. We also combine the two modules and analyze the response in the simulation. In real flights, the classifier (FDI module) responds to the fault in 2-5 sensor data samples (at a 60ms rate per sample) and has a high true positive rate of 92.6%. Also, in real flights, the performance of the FTC module is measured in terms of tracking error, percentage overshoot, and settling time. Integration of the FDI and FTC modules is performed and simulation results are presented showing satisfactory operation of each of these modules with guaranteed stable flight.

Abstract

Automatic Speech Recognition (ASR) systems are increasingly prevalent in our daily lives, with commercial applications such as Siri, Alexa, and Google Assistant. However, the focus of these systems has been largely angled towards English, leaving a considerable portion of non-English speakers underserved. This is particularly evident in India, a linguistically diverse country with many languages classified as low-resource in the context of ASR due to the scarcity of annotated speech data. This thesis aims to bridge this gap, focusing on enhancing ASR systems for Indian languages using deep learning methodologies. India is a land of language diversity. There are approximately 2000 languages spoken around, and among those officially registered are 23. Of those, very few have ASR capability. This is because building an ASR system requires thousands of hours of annotated speech data, a vast amount of text, and a lexicon that can span all the words in the languages. The necessity for a comprehensive presence in the diverse Indian markets demands the development of multilingual Automatic Speech Recognition (ASR) systems. It’s a common scenario where ASR systems for Indian languages have to be implemented in low-resourced contexts. Furthermore, the complexity of the linguistic landscape is amplified due to the high prevalence of bilingualism in the Indian population, leading to frequent instances of code-switching and linguistic borrowing between languages. Operating concurrent ASR systems that can handle code-switching in the Indian context presents a considerable challenge. This predicament has spurred our research endeavors, driving us to focus on constructing a large corpus for one language and leveraging its phonetic space on other language families in monolingual and multilingual ASR scenarios. This thesis incorporates a crowd-sourcing strategy to collect an extensive speech corpus, particularly for Telugu. Using this approach, around 2000 hours of Telugu speech data, capturing regional variations through three modes: spontaneous, conversational, and read, under various background conditions, has been collected. This data served as a foundation for developing and evaluating neural network architectures tailored to the characteristics of Indian languages. We also explored the potential of self-supervised learning to understand and enhance learned representations, fine-tuning them to suit different language families and data sizes. This approach led to insights into the shared phonetic space among Indian languages, allowing for the development of a multilingual ASR system utilizing a joint acoustic model approach. These studies mark a significant stride towards overcoming the challenges of multilingualism in the Indian context, setting a path for creating more inclusive and effective ASR systems. The research findings presented in this thesis not only contribute towards building efficient and accurate ASR systems for low-resource Indian languages but also underscore the power of deep learning approaches in linguistic technology. It is our hope that this work will motivate and aid further research in this direction, promoting linguistic diversity and broadening access to information and communication technologies for speakers of low-resource languages.

Abstract

In this thesis, we try to tackle the problem of erroneous English-Hindi machine translation (MT) outputs due to the presence of the Verb Phrase Ellipsis (VPE) in English. The phenomenon of VPE is prominent in spoken English, and the antecedent to the ellipsis can come from previous sentences in a conversation as well. MT systems translate sentences as a whole and ignore the contextual information from the previous sentences. For these two reasons, spoken English-Hindi translations suffer. We approached this problem by manually annotating 1200 two-person conversations that contain VPE and by studying how their resolution affects the translation qualities. Based on this analysis, we designed a rule-based system for the detection and resolution of VPE in English with the goal of improving their subsequent Hindi translation qualities. Our rule-based system is capable of the following: 1) Detection of VPE, 2) Resolution of Elided Head verb, 3) Resolution of Elided Head verb’s children, 4) Resolution of non-verbal predicates of a copula or a ’be’ main verb, 5) Modifying original sentence in the conversation with the resolved verb phrase. In order to assess the scalability of our rule-based system, we also tested the system’s performance on VPE datasets outside of our annotated data. The system’s performance was also tested on nonconversational data in this research. The scalability test allowed us to utilize our rule-based system for the purpose of data augmentation as well. The augmented data was utilized for the training of a Transformer-Based Seq2Seq generation model. This model helped us investigate whether a sequenceto-sequence model can be designed to handle instances of conversational English VPE, which could be utilized as a pre-processing step before carrying out English-Hindi Machine Translation. Furthermore, a comparative study between the rule-based and the Transformer-based model was carried out as a part of the research supporting this thesis.