Abstract

Always-on agents are those agents which are like daemon programs that are always on and can do tasks as and when they arrive. These agents are idle when there is no task assigned to them. Further, those agents that work on a task also wait for some event or task completion, and hence, are also idle for short durations while executing the tasks. The question arises what should the agent be doing when it is idle. In this thesis, we conduct an empirical analysis to show improved decision-making capabilities by agents when they exploit their idle time. We analyze scenarios where (i) the agents analyze their past tasks regarding decisions taken and their impact and where (ii) the agents cooperate with other agents to improve their decision-making. Performance improvement can be measured by considering an increase in success rate, avoiding strategies that may not work, quicker decision making, etc. While executing a task, our always-on agent stores pertinent details of the task done in a database, such as decisions taken, paths of execution of the task, goodness of them, etc. The agent uses different strategies to use this stored knowledge. We present and evaluate three strategies that always-on agents can use (i) Frequent Decision Strategy (FDS) - the agent stores the prior executions and their frequency of success and failure, repeats the most frequent successful decision taken during prior executions of the task, (ii) Analyzed Decision Strategy (ADS) - the agent analyses prior executions that were successful or not, stores in the database the goodness of various alternatives and chooses the best alternative and (iii) Online Analysis Decision Strategy (OADS) - the always-on agent while executing its task, during its idle time analyses the possible future situations and prepares the list of best possible decisions that can be taken in future. The FDS and ADS are used when the agent is not doing any task and is off-line. In contrast, OADS generates new mock tasks to consider possible alternative task execution situations to expand its decision-making scenarios while having a task at hand. We conduct our empirical study on always-on agents playing Connect-4 games to check the viability and improved decision-making. We also present idle time analysis in always-on cooperative agents. We propose a Multi-agent Framework that always-on cooperative agents can use in their idle time to improve decision-making. We formulate scenarios in which the agent has a rationale to cooperate with other agents (i) Exception Handling - the agent cooperates to handle situations in which it does not know what to do, (ii) Confidence Boosting - the agent cooperates to boost confidence in actions it wants to take, (iii) To obtain Different Experiences - the agent cooperates to obtain different experiences. We conduct an empirical study on always-on cooperative agents using the Multi-agent Framework and show improved decision-making.

Abstract

Spontaneous speech is a particular type of speech setting where a speaker speaks without preparing in advance. This makes the speaker think about what to say on the spot, formulate the utterances and then produce the speech. Such a setting often leads to abrupt breaks or discontinuities in the normal conversation flow called disfluencies. Disfluencies can provide information regarding the speaking style, speaker identity and language fluency, which can be useful for several speech-based applications. For automatic speech recognition (ASR) systems, the presence of these disfluencies leads to a higher word error rate, since most ASR systems are developed on non-spontaneous read speech data. Thus, the detection of disfluencies in spontaneous speech becomes an essential task for many applications. For training any machine learning system, one need data. In this thesis, we introduce IIITH-IED dataset for disfluencies in spontaneous lecture-mode speech, and then use it to develop frame-level automatic disfluency detection systems. Finally, we propose a transfer learning method to detect disfluencies in spontaneous lecture mode speech using three frame-level automatic disfluency detection systems trained on stuttered speech. Compared to baseline systems, proposed method gives an average improvement of 2.25% across all disfluencies and all detection systems.

Abstract

The extent of damage in a structure can be estimated numerically using Nonlinear analysis (NA). On the basis of the applied forces, NA can further be classified as Static Non-linear (Pushover Analysis) and Dynamic Non-linear (Time History Analysis). The response of a structure can be predicted effectively using Non-linear analysis, provided adequate assumptions are made. The efficacy of Non-linear analysis depends on the accurate assumptions pertaining to types, locations and pattern of non-linear hinges. Various hinge definitions that are generally used for NA of a structure includes Axial, Shear, Flexural or Torsional force-deformation relationships. However, these force-deformations are seldom mutually exclusive, and can be coupled to achieve a better numerical simulation of actual behaviour, and failure mechanism. The most common combination of hinge definition applied for NA of 2D frames is P-M2 hinges (P-M2-M3 in case of Space Frames). As per the literature, the three most common modes of failure in low and medium rise structures are Shear failure, Flexural failure, and a combination thereof. This observation emphasizes the need to consider the combination of shear and flexural hinge definitions during analysis, to better estimate the behaviour and effectively predict the failure pattern of a structure. The literature also suggests an inversely proportional relationship between the shear and moment capacity of a section beyond first yield. The prevalent practice in a typical design office requires an engineer to define a flexural hinge (in terms of moment-rotation/curvature relationship) for a section, and assigning a value for ultimate shear strength causing brittle failure, and thereby neglecting the effect of shear on the non-linear behaviour of the section. These observations form the primary objective of the work presented here, which is to develop a Shear-Moment interaction behaviour for a Reinforced Concrete (RC) section. Among the commonly constructed RC structures, Open Ground Storey (OGS) structures have swiftly become the most preferred type of RC buildings, primarily owing to large open spaces on the ground level. However, structural engineers deem OGS as a form of vertical irregularity, and advice special care while designing, to prevent collapse, especially in seismic prone areas. The literature consists of a plethora work on modelling of infill panels, especially in case of an OGS building. Various models exist to effectively simulate the effect of infill wall panels, the most common of which are single strut model, double struct model and three strut model. However, in a typical design office, buildings are generally analysed as bare frames, thereby neglecting the effect of infill panels, and leading to incorrect estimation of behaviour of the building. This practice, when employed for an OGS building leads to further deviation of analysis results from the actual behaviour. Furthermore, when designed as bare frames, the buildings having OGS exhibit higher strength as well as stiffness on the storeys above the ground storey. Since the OGS is both a soft and a weak storey, it will incur higher stresses, leading to significant damage. The damage sustained by the structural members in the OGS depends primarily on the ductility of columns, relative lateral stiffness and strength with respect to the adjacent upper storeys. Further, the height of the building having an OGS has a significant effect on its behaviour. The various factors governing the behaviour of a building having OGS can be combined together to obtain a Amplification Factor (AF), that can be used as a multiplication factor to modify the demand on the OGS. This forms another important objective of the present work. The AF will primarily depend upon: 1) Material property of masonry, 2) Design type of RC building, 3) Percentage of openings present in upper floors, and 4) Aspect ratio of the building. In present study, an attempt is made to determine the shear capacity of reinforced concrete section with respect to different deformation levels and to understand the effect of shear-moment interaction of reinforced concrete section in nonlinear zone while performing nonlinear analysis of the structure. Alternatively shear hinge of V-Ø type is proposed in place of existing one value of shear capacity and its effect on nonlinear response of the infilled RC structure because of change in shear failure mechanism is studied. Also an attempt is made to compute design amplification factor (DAF) for safety of OGS during seismic events considering different parameters aforementioned mentioned above and numerical correlation of same with DAF is done using linear regression analysis.

Abstract

A major customer demand in the software industry today is quality at speed. Systems and Software Development Lifecycle processes (SDLC) are central to the work in IT Organizations that guide product and service development and act as vehicles for building the engineering deliverables (ISO/IEC 12207, 2017; ISO/IEC 15288, 2015). Despite adopting standard processes, IT organizations continue to face enormous quality-related challenges in developing engineering deliverables primarily because the standards outline “what” tasks to be carried out but not the “how” part of developing solutions. There is a lot of reliance on the knowledge and expertise of the people involved in the SDLC to build quality solutions for the stated problem. More specifically, early SDLC phases such as proposals, requirements are document-centric and manually intensive. The essential knowledge related to the domain and solutions is buried in various kinds of documents in shared repositories. Subject Matter Experts (SME) refer to and reuse content from these documents while preparing solutions for a given problem in early SDLC. The only technological help is often a “Ctrl-F” on individual documents to search and pick up relevant information. Even though downstream SDLC phases like coding and testing have long been recognized as focus areas for productivity, and automation exists in terms of code generation and test automation, the technology and automation paradigms are under-exploited in early SDLC phases. The responsibility for solution development during the early SDLC phases lies entirely with the SMEs, and the whole process is entirely manual. There is a huge gap in process infrastructure support for the early SDLC. AI in software development lifecycle is an emerging area of research with a focus on applying intelligent tools and techniques to improve/ accelerate/ disrupt the life cycle. As part of the “AI in SDLC” paradigm, technologies such as natural language processing (NLP), rule-based reasoning are widely explored. This thesis investigates how specific processes in early SDLC can be digitally reimagined to facilitate the transformation from a manual approach to an automated approach. This thesis proposes a generic intelligent Systems/ software Process INfrastructure (iSPIN) framework focusing on extracting solution knowledge into machine-processable models and building intelligent recommender systems for a given problem in the early phases of SDLC. The key contributions of this research are: i) Literature and Field Studies on Quality Models and SDLC Practices to understand the quality challenges and establish the quality gaps in SDLC ii) iSPIN framework for building intelligent technology infrastructure for early SDLC phases iii) A generic SDLC System meta-meta-model (SDLC_MMM) for solution development iv) Case Study1: iSPIN implementation and evaluation for “Request for proposal (RFP) response” process v) Case Study2: iSPIN implementation and evaluation for “Requirements” process The major contribution of this work is the iSPIN framework with crucial building blocks to build an intelligent process infrastructure. A set of technology enablers are proposed comprising an SDLC System meta-meta-model, AI-based knowledge extraction, and intelligent solution recommender. The SDLC system meta-meta-model (SDLC_MMM) captures the essential SDLC process elements and their interactions. AI-based knowledge extraction parses the natural language (NL) documents and automatically extracts key knowledge elements with associated knowledge classification and context tagging. Finally, an intelligent solution recommender analyzes the specific problem in the SDLC phase, determines applicable concerns, recommends relevant knowledge, and automatically composes a machine-first “SDLC output” document as per the required document format providing significant productivity benefits. iSPIN is implemented as a proof of concept (PoC) technology and validated as part of two early SDLC processes for the below use cases - i) For automatically generating a Response to a clientsupplied Request for Proposal (RFP) / Request for Information (RFI) response ii) For generating context-sensitive user stories from diverse Requirements Specifications. For the first use case, The efficacy of the approach has been demonstrated on a set of RFP/RFIs. Based on the case study implementation, it has been observed that for any RFx received from the client, iSPIN can potentially generate 50%-70% of RFP/RFI response automatically. This capability of “Machine First RFP/RFI Response” is a huge value-add for the Presales teams. For the second use case, demonstrated results on two products and 90 requirement specification documents. The model extraction efficiency is very good, with 97% accuracy. While iSPIN has been conceived, developed, and validated for the early SDLC processes, the guiding principles, SDLC_MMM, and the iSPIN framework

Abstract

In distributed storage systems, two kinds of models are used for repairing in case of multiple erasures. In centralized model, a central node downloads the information from all the helper nodes and reconstructs the failed nodes and then sends reconstructed data to corresponding node. While in cooperative model, repair happens in two rounds and each reconstructing node downloads some information from helper nodes and other reconstructing nodes. In rack-aware distributed storage systems, there is no cost associated with transferring symbols within a rack. Hence, the repair bandwidth will only take into account cross-rack transfer. Rack-aware regenerating codes for the case of single node failures have been studied and their repair bandwidth tradeoff characterized. In this thesis, we consider the framework of rack-aware cooperative regenerating codes for the case of multiple node failures where the node failures are uniformly distributed among a certain number of racks. We characterize the storage-repair bandwidth tradeoff as well as derive the minimum storage and minimum repair bandwidth points of the tradeoff. We make use of both centralized and cooperative model to arrive at the storage-bandwidth tradeoff and the minimum storage/bandwidth points. We also provide constructions of minimum bandwidth rack-aware cooperative regenerating codes for all parameters. In next part of the thesis, we present our implementation of Clay Codes in Hadoop. Hadoop is a software platform which provides a framework for distributed storage. Hadoop file system can store data according to different erasure coding algorithms. Some algorithms like replication, xor coding and reed-solomon codes are already implemented in Hadoop source code. Clay Codes are vector codes where data within a node is further divided into chunks. Clay Codes are a type of high-rate MSR codes. For our work, we have tried the implementation of clay codes and discussed the challenges faced

Abstract

Unmanned aerial vehicles (UAVs) are considered to be among the most advanced robotics technologies available today. They are capable of supporting and transporting sensors, communication equipment, and a variety of other payloads. UAVs are cost-effective, and with better sensing technology, they are being utilized by a wide variety of enterprises, both military and commercial. They have been acknowledged in recent years for their ability to perform a variety of tasks, particularly in the military domain like surveillance, convoy protection, border patrol. One of the key aspects involved in these military applications is target tracking. The existing literature on target tracking is primarily focused on tracking a single target. In this thesis, a Model Predictive Control (MPC) based algorithm is proposed for tracking multiple ground target swarms of unmanned aerial vehicles (UAVs). Swarm of UAVs have distinct advantages when compared to a swarm of unmanned aerial vehicles (UGVs). When used for applications like multi-target tracking with a sensor like a camera, tracking through aerial view avoids issues like loss of observability which is typically associated with monocular vision-based target tracking from the ground. Another advantage of UAVs over UGVs is that the tracking in case of UAVs is simple while tracking through UGVs due to obstacles and the restricted environment becomes complex. There are two primary types of UAVs, namely fixed-wing aircraft(FWA) and multi-rotor UAVs. FWA because of their ability to cover longer distances at higher speeds, are preferred in military applications, but they come with a disadvantage that they come with lower bound on speed and non-zero-turn radius. Multi-target tracking through UAVs has numerous difficulties, owing to the maneuverability of target, atmospheric disturbances. MPC has been shown to be an efficient method for process control, tracking, path planning, and the ability to handle disturbances well, among other applications. Future commands with a short time horizon are determined with high precision using future prediction and optimization techniques. The primary argument for choosing MPC over other techniques is its ability to adequately account for constraints, allowing all processes to run at their maximum possible performance. All UAVs here belong to the fixed-wing aircraft category having flight velocity, climb rate, and turn rate limits. Each UAV is equipped with a downward-facing camera for the purpose of detecting and tracking the target. Two scenarios are studied in which the total number of UAVs equals the total number of targets in the first scenario. In the second situation, the number of UAVs is less than the number of targets, resulting in a conservative solution in which the objective is to maximize the average time duration that targets are within the FOV of any of the UAV’s cameras. To relate the hyperparameters utilized in MPC to mission efficiency, a data-driven Gaussian process (GP) model is created. The GP model provides black-box identification of non-linear dynamic systems using a probabilistic non-parametric modeling technique. Gaussian processes can identify regions of the input space where prediction accuracy is low due to a lack of data or its complexity by offering a greater degree of variation around the anticipated mean. Bayesian optimization is used to determine the MPC hyperparameters that maximize mission efficiency by treating the algorithm’s generalization performance as a sample from a Gaussian process. The tractable posterior distribution generated by the GP enables the most efficient use of data from previous trials, allowing for the best decisions on which parameters to test next. Both situations are numerically simulated using a distributed MPC formulation technique. Apart from numerical simulations, the following performance comparisons are made by employing constraints in different ways: first, between decentralized and centralized approaches via error vs. time plots for individual UAVs, and then between computing efficiency and root mean square error for different prediction horizons. Second, between the proposed method and greedy approach for both centralized and decentralized MPC implementation

Abstract

Engineering proteins to have desired properties by mutating amino acids at specific sites is commonplace. Such engineered proteins must be stable to function. Experimental methods used to determine stability at throughputs required to scan the protein sequence space thoroughly are laborious. To this end, many machine learning based methods have been developed to predict thermodynamic stability changes upon mutation. They were trained using small and biased datasets. The methods have been previously evaluated for symmetric consistency by testing with hypothetical reverse mutations, and many fail the test. In this work, we propose transitive data augmentation, evaluating transitive consistency with our new S transitive dataset, and a new machine learning based method, first of its kind, that incorporates both symmetric and transitive properties into the architecture. Our method, called SCONES, is an interpretable neural network that predicts small relative protein stability changes for missense mutations that do not significantly alter the structure. It estimates a residue’s contributions towards protein stability (∆G) in its local structural environment, and the difference between independently predicted contributions of the reference and mutant residues is reported as ∆∆G. We show that this self-consistent machine learning architecture is immune to many common biases in datasets, relies less on data than existing methods, is robust to overfitting, and can explain a substantial portion of the variance in experimental data.

Abstract

Automated segmentation of cortical and non-cortical human brain structures has been hitherto approached using nonrigid registration followed by label fusion. We propose an alternative approach for this using a convolutional neural network (CNN) which classifies a voxel into one of many structures. Four different kinds of two-dimensional and three-dimensional intensity patches are extracted for each voxel, providing local and global (context) information to the CNN. The proposed approach is evaluated on five different publicly available datasets which differ in the number of labels per volume. The obtained mean Dice coefficient varied according to the number of labels, for example, it is 0.844 0.031 and 0.743 0.019 for datasets with the least (32) and the most (134) number of labels, respectively. These figures are marginally better or on par with those obtained with the current state-of-the-art methods on nearly all datasets, at a reduced computational time. The consistently good performance of the proposed method across datasets and no requirement for registration make it attractive for many applications where reduced computational time is necessary. Feature-based registration has been popular with a variety of features ranging from voxel intensity to Self-Similarity Context (SSC). In the second part of the thesis, we examine the question of how features learnt using various Deep Learning (DL) frameworks for segmentation can be used for deformable registration and whether this feature learning is necessary or not. We investigate the use of features learned by different DL methods in the current state-of-the-art discrete registration framework and analyze its performance on 2 publicly available datasets. We draw insights about the type of DL framework useful for feature learning. We consider the impact, if any, of the complexity of different DL models and brain parcellation methods on the performance of discrete registration. Our results indicate that the registration performance with DL features and SSC are comparable and stable across datasets whereas this does not hold for low level features.

Abstract

Internet of Vehicles (IoV) is an extension of Vehicle-to-Vehicle (V2V) communication network. IoV is a connected adhoc network (Vehicular Ad Hoc Network (VANET)), where each vehicle in the network is a node, connected to other vehicles and also to the public Internet. IoV helps in enhancing driving aids with the help of vehicle’s Artificial Intelligence (AI), awareness of other vehicles and their actions. The autonomous vehicles can instantaneously communicate with other vehicles surrounding them. Since the communication among various entities involved in the IoV environment is via open channel (e.g. vehicles, pedestrians, fleet management systems, and road-side infrastructure), it gives an opportunity to a passive/active adversary to intercept, modify, delete or even insert fake information during communication. It is then a serious concern for the vehicles users to determine whether the received information is genuine. Many security protocols have addressed this issue by adding authentication mechanism. Authentication of sender is a process of verifying his/her identity claimed by or for a system entity. Authentication of the messages ensures that the messages which is flowing in the network is unintercepted and unforged. Mutual authentication is a type of authentication which allows the involved entities simultaneously authenticate each other in order to establish a secure communication between them via computed session key. In this thesis, we study the IoV paradigm and its comparison and evolution over VANETs. We present a taxonomy of security protocols designed for IoV security. We address the threats and attacks in IoV environment. In particular, we focus on various authentication protocols in IoV. We propose authentication schemes, to accomplish mutual authentication, conditional privacy preserving, batch authentication and blockchain based authentication among the involved entities in the IoV environment for secure communication. For every proposed scheme, a detailed comparative analysis among various state-of-art authentication protocols proposed in the related IoV environment is provided to show their effectiveness as well as security and functionality features. The first study presents a new mutual authentication and key agreement protocol in Internet of Vehicles-enabled Intelligent Transportation System (ITS), using the elliptic curve cryptography (ECC) technique. The proposed protocol deals with the challenges during open communication among entities in IoV environment. An open communication can be targeted by the adversary to eavesdrop, modify, insert fabricated (or malicious) messages, or delete any data-in-transit; thus, resulting in replay, impersonation, man-in-the-middle, privileged-insider, and other related attacks. In addition to providing security, the proposed scheme also achieves anonymity and untraceability. It provides a mutual authentication between vehicle and vehicle and also between a vehicle and a road side unit (RSU). Using both formal and informal security analysis, as well as formal security verification using an Automated Validation of Internet Security Protocols and Applications (AVISPA) tool, the proposed scheme is proved to be secured against several known attacks in an IoV-enabled ITS environment. Furthermore, a detailed comparative analysis shows that the proposed scheme has low communication and computational overheads, and offers better security and functionality attributes in comparison to other competing schemes. We also evaluate the performance of the proposed scheme using NS2. In our second study, we design a new conditional privacy preserving batch verificationbased authentication mechanism in the IoV environment using ECC technique, where a vehicle can authenticate its neighbor vehicle and also an RSU can authenticate its nearby vehicles in a batch. The proposed scheme incurs low computation cost as it is implements batch authentication. The proposed scheme is shown to be highly secure against a passive/active adversary through various security analysis, such as random oracle based formal security, formal security verification via automated simulation tool (AVISPA), and also informal security analysis. An exhaustive comparative analysis reveals that the proposed scheme offers better security and functionality attributes, when compared with the relevant schemes. Finally, in the third study, we focus on designing a novel blockchain-enabled batch authentication scheme in Artificial Intelligence (AI)-envisioned IoV-based smart city deployment. We collaborate mutual authentication, batch authentication and blockchaining to produce an efficient scheme. The vehicles in IoV can be used to opportunistically gather and distribute the data in a smart city environment. However, at the same time, various security threats arise due to insecure communication happening among various entities due to lack of trust. Moreover with the increase in number of vehicl

Abstract

In the first part of this thesis, we analyze a multi-cloud marketplace model consisting of brokers with multi-unit demands and cloud providers with multi-unit supplies. Using tools from cooperative game theory we show the existence of a core allocation which also maximizes the market surplus. We do this by finding an optimal coalition structure and formulate an integer linear programming problem to maximize the overall market surplus. The optimal coalition lies in the core which gives it stability and no incentive to deviate to any other sub-coalition. We then study our model in the real-world scenario in which the brokers don’t reveal their valuations for the cloud providers. For this, we use the auction algorithm due to Bertsekas [2] which leads to a core allocation without requiring the brokers to reveal their utilities. In the second part of this thesis, we study how an oligopolist influences the coalition structure in federated cloud markets. We consider two price offering strategies for an oligopolist: non-adaptive and adaptive. In the non-adaptive strategy, an oligopolist makes a price offer to all the cloud providers simultaneously. It can be noted that the oligopolist can buy-out all the cloud providers by making a price offer which is equal to a core allocation and the total price offer made by the oligopolist is equal to the value of the grand coalition. In the adaptive strategy, the oligopolist approaches the cloud providers one after another in a sequential manner. We show that by using the adaptive strategy, the oligopolist can buy-out all the cloud providers at a total price offer which is less than that of the non-adaptive strategy