Abstract

This research explores Bayesian methods for predicting model uncertainty in remote sensing data, focusing on two distinct approaches: Bayesian linear regression and a novel Bayesian Nonlinear Neural Network (BN3) model. Both methods utilize SENTINEL-2 Normalized Difference Vegetation Index (NDVI) data from agricultural fields in Uttar Pradesh and Madhya Pradesh, India, for time series predictions. Bayesian linear regression, a probabilistic deep learning approach, evaluates model uncertainty through prior and posterior probability parameters using Bayesian inference. The analysis reveals high model uncertainty in predicted NDVI values, which diminishes as the data volume increases. This reduction in uncertainty is accompanied by sharper posterior density, indicating reduced variance. Regression analysis with Gaussian distribution further validates this trend, confirming that greater data availability leads to improved model certainty. Complementing this approach, the study develops the BN3 model to address model uncertainty in neural networks. BN3 employs probabilistic modeling with variational hidden layers and output layers characterized by Gaussian distributions for variational posteriors. It assesses model uncertainty using dense variational layers with continuous, differentiable nonlinear activation functions and varying degrees of freedom. The study evaluates the goodness of fit for nonlinear models, including polynomials and neural networks, using RMSE as a loss function. Comparison with Deep Ensemble methods highlights the BN3 models effectiveness, particularly in reducing uncertainty for cubic models. Visual analyses demonstrate BN3s consistent performance in minimizing model uncertainty across datasets, even with varying data availability. These findings demonstrate BN3s potential as a robust tool for improving uncertainty quantification in decision support systems.

Abstract

The widespread use of artificial intelligence has resulted in significant advancements in the field of Natural Language Processing, with text summarization and question answering models being some of the most sought-after applications. By understanding the Causal Relationships between Events within a text, we can gain valuable insights into how Events influence each other, leading to more accurate and sophisticated model outcomes. This thesis is an attempt towards improving Event Causality through various ways. The thesis commences with a comprehensive examination of Events and States, outlining their various classifications and the relationships between them. This foundational discussion establishes the context for the exploration of Event Causality. It then delves into existing research on Event Causality identification highlighting areas for potential contribution Recognizing the scarcity of datasets focused on Event Causality in the Hindi language, this thesis introduces the Hindi Causal TimeBank. This dataset is built upon the existing Hindi TimeBank containing about 1000 news articles and can serve as a crucial resource for developing and training Event Causality models for Hindi-language text. The thesis elaborates on the different types of Causal Relations annotated within the dataset. The primary focus of this thesis lies in the advancement of Event Causality identification. Specifically, it explores the use of graph-based models (Graph Attention Networks (GAT)). GATs are well-suited for representing complex relationships between entities and Events. This thesis presents a series of experiments designed to improve results on Event Causality identification, aiming to outperform the current state-of-the-art methods. The performance of these models is evaluated across a range of datasets, including the newly developed Hindi Causal TimeBank. This thesis makes a significant contribution to the field of Event Causality. Improved Event Causality understanding has implications for information retrieval, machine translation, and knowledge graph construction. The introduction of the Hindi Causal TimeBank further expands the scope of this research into a crucial and understudied language.

Abstract

Enhancing quadrotor efficiency through adaptive control is essential for addressing the crit- ical need for Fault-Tolerant Control (FTC) in the presence of operational uncertainties and component inefficiencies. Conventional adaptive FTC strategies typically assume that uncer- tainties are bounded by a constant known a priori. However, this assumption is inadequate when dealing with imprecise inertial system parameters, leading to state-dependent uncertain- ties that deviate from this model. When left unaddressed, such uncertainties can result in insta- bility, particularly under actuator faults, posing significant risks to the reliability of quadrotor systems. To overcome these challenges, the proposed adaptive FTC method effectively mitigates ac- tuator faults and addresses unknown state-dependent uncertainties through the implementation of carefully designed adaptive laws. This approach incorporates real-time fault detection and dynamic control allocation, which together prevent overly conservative control applications, ensuring that the quadrotor maintains optimal performance even under adverse conditions. By dynamically adjusting control efforts based on real-time data, the system is capable of respond- ing robustly to varying fault scenarios, thereby enhancing overall operational resilience. The stability of the closed-loop system is rigorously analyzed using Lyapunov theory, which provides a strong theoretical foundation for the proposed framework. Analytical studies con- firm that the adaptive FTC maintains stability across a wide range of operational scenarios, en- suring the quadrotor’s reliability. The effectiveness of this solution is further validated through comprehensive simulations on a realistic quadrotor model, demonstrating significant improve- ments over existing methods in terms of both stability and efficiency. This research highlights the importance of adaptive control strategies in maintaining reliable and safe quadrotor operations, setting the stage for future advancements in UAV fault tolerance and control systems. By addressing the complexities of state-dependent uncertainties and actu- ator faults, the study contributes to the broader field of autonomous aerial vehicles, ultimately enhancing the safety and effectiveness of UAV deployments in various applications.

Abstract

Music has therapeutic potential for individuals with Autism Spectrum Disorder (ASD). Traditional music therapy often lacks personalization, with music selected by therapists rather than tailored to individual preferences. Current studies on ASD individuals’ music preferences often overlook cultural inclusivity and the impact of lyrics. Moreover, existing recommendation systems do not consider music preferences tailored for therapeutic goals. This thesis explores how music aids emotional regulation, cognitive functioning, and social interaction in individuals with ASD, emphasizing cultural inclusivity. The study includes analyses of music specific community for neurotypical individuals (NT) and ASD communities on Reddit, cross-cultural comparisons between Non-Indian and Indian populations, a pilot study with Indian children with ASD, and the development of a personalized Music Recommendation System. Reddit analysis revealed diverse music preferences, underscoring the need for personalized interventions. Cross-cultural comparisons showed non-Indian participants preferred Rock and Western Classical music, while Indian participants favored Indian film and Devotional music, reflecting cultural influences on emotional expression. The pilot study with Indian children identified diverse music preferences, including Indian Devotional music, Nursery rhymes, and Indian film music. Listening to preferred music indicated positive behavioral changes, such as reduced anxiety and improved social interaction, highlighting music’s therapeutic potential. The GOAL (Genre, Optimization, Acoustic Features) based Music Recommendation System offers personalized suggestions based on genres, acoustic features, and lyrical themes relevant to emotional regulation, cognitive functioning, and social skills. By categorizing goals into emotional, cognitive, and social functions, the system aligns with diverse ASD needs and provides personalized music recommendations using similarity metrics. Evaluation metrics indicated the model accurately recommends relevant music, achieving a high proportion of liked music (precision) and capturing a significant portion of preferred music (recall). Behavioral changes showed effectiveness in mood management, focus, sensory load regulation, and sleep regularity, with mixed results in communication and social skills. In conclusion, this thesis underscores the importance of recognizing and accommodating ASD individuals’ music preferences in therapy. Despite limitations such as small sample sizes and subjective reports, findings offer insights into music’s therapeutic potential for ASD. Future research should include larger, diverse samples and longitudinal studies to enhance the proposed system’s effectiveness.

Abstract

Scene text conveys important information to drivers and pedestrians, serving as an indispensable means of communication in various environments. Scene text contains information regarding the speed limit, route information, rest stops, and exits, among other important information. It is important for drivers and passengers to understand this information for a safe and efficient journey. However, outdoor scenes are cluttered with text, distracting drivers and making it hard to focus on what matters, potentially compromising their ability to focus on essential details and navigate safely. Recognising scene text in motion aggravates this challenge, as textual cues transiently appear and necessitate early detection at a distance. Driving scenarios introduce additional complexities, including occlusions, motion blur, perspective distortions, and varying text sizes, further complicating scene text understanding. In this thesis, we look at improving scene text understanding in diving scenarios through video question answering and analysing the present state of scene text detection, recognition and tracking: (i) We introduce a new video questions answering task and dataset that requires an understanding of text and road signs in driving videos to answer the questions. (ii) We look at the current state of scene text detection, tracking and recognition in the driving domain through the RoadText-1K competition. (iii) We explore detection and recognition in special cases of occlusions, a common yet under-explored complication in real-world driving scenarios. By focusing on these areas, the thesis contributes to advancing scene text analysis methodologies, offering insights and solutions that are imperative for developing more intelligent and responsive driver assistance systems.

Abstract

This thesis addresses the critical challenge of monitoring water flow to support efficient water resource management and conservation. It presents an innovative approach to developing a sustainable water management system through an Internet of Things (IoT)-based solution. The primary objective is to retrofit existing analog water meters using affordable, off-the-shelf electronic components, transforming them into smart devices capable of real-time data collection and analysis. A distinctive feature of this system is its use of edge computation, which allows meter readings to be extracted directly from images captured by an onboard camera. This local processing approach minimizes latency and reduces the need for continuous cloud communication, enhancing system responsiveness and reducing data transmission costs. Energy efficiency and sustainability are critical components of this design. To support autonomous operation, the system is powered by solar-charged batteries, with strategies implemented to preserve charge-discharge cycles even in challenging environmental conditions. Additionally, an automated broadcasting technique facilitates firmware updates across multiple devices simultaneously, reducing the need for manual intervention and enhancing overall maintenance efficiency. The effectiveness and reliability of the system were validated through an extensive field deployment involving 15 smart water meter nodes installed across various locations at IIIT-H campus and monitored continuously for over 2.5 years. This long-term deployment provided valuable real-world data, allowing for detailed characterization of water usage patterns across different environments. The stable operation of these nodes over a prolonged period demonstrated the robustness and durability of the design, supporting its potential for broader application in diverse settings. An essential feature of the system is its energy-harvesting capability, which minimizes reliance on external power sources and significantly reduces maintenance requirements. This feature aligns with sustainable energy practices and promotes an eco-friendly approach to water resource management. By integrating efficient data acquisition, local processing, and cloud-based storage and analysis, this IoTbased solution offers a comprehensive, scalable, and sustainable approach to advancing water resource management and supporting global conservation efforts.

Abstract

Integrating Artificial Intelligence (AI) into drug design has revolutionized the field at various levels, ranging from simple molecular property prediction tasks to complex de novo drug design tasks. This is still a relatively young field and is rapidly evolving. In this thesis, we identify various challenges AI models face due to data constraints and analyze hidden biases in these methods; we improve upon existing molecular generation models for aligning the drug design to desired complex properties; we propose a novel framework for structure elucidation from molecular spectra. The second chapter explores various biases in protein-ligand datasets that affect the performance of deep learning models in binding affinity prediction and virtual screening. It demonstrates how random data splitting leads to overly optimistic results and introduces more nuanced data splitting methods that account for sequence, pocket structure, and protein-ligand interaction similarities. The research identifies biases in these datasets related to protein-only and ligand-only information. It proposes improvements in dataset construction and model design to enhance the generalizability and accuracy of AI-based protein-ligand scoring functions. In the third chapter, we develop and refine Molecular generative models for more accurate conditional generation. We propose optimization methods aligning the generated molecules’ properties with the desired ones. The fourth chapter introduces the Spectra and Molecule Encoder Network (SMEN) for scoring molecules against target spectra. The model aims to aid library ranking tasks for Infrared (IR) Spectrumbased structure elucidation. SMEN learns spectral and molecular embeddings, accurately representing both entities in a high-dimensional latent space. Additionally, the SMILES Decoder (SD) generates SMILES strings from the latent space, enabling accurate one-shot predictions from IR spectra to molecule structures and showcasing promising accuracies for rapid, precise structure elucidation. In summary, This thesis explores the potential of ML models to predict binding affinities and generate molecular structures with desired properties. It begins by identifying and addressing latent biases in popular protein-ligand datasets, affecting binding affinity predictions’ accuracy. This research introduces more refined data-splitting methods that enhance model generalizability by analyzing biases in sequence-based and structure-based datasets. The thesis then presents a generative model for generating molecules with specific characteristics, such as binding affinity, and further optimizes the model’s ability to produce molecules that meet predefined criteria. Lastly, a contrastive learning framework is developed for generating molecular structures from infrared spectra, demonstrating the model’s versatility in handling various molecular generation tasks. Each chapter addresses foundational issues, setting the stage for future innovations in the field.

Abstract

This research delves into the application of advanced semantic segmentation techniques to automate facade inspection, a critical component in the maintenance and safety assessment of urban infrastruc- tures. Central to the study is the WALL-E dataset, a comprehensive collection specifically curated to encompass a wide range of facade defects, including cracks, stains, and delamination. This dataset is designed to challenge and refine the capabilities of deep learning models by exposing them to a diverse array of defect types and complex architectural features, thereby mimicking the variety of conditions encountered in real-world settings. The training methodology employed in this study leverages the strengths of the DeepLabV3+ model, compared against other leading models such as UNet, PSPNet, and FCN. Each model was rigorously trained and evaluated on their ability to accurately segment defects across different types of building materials and architectural styles. DeepLabV3+ emerged as the standout model due to its sophisticated architecture that combines atrous convolution with an encoder-decoder structure, enhancing its ability to capture detailed spatial information and accurately delineate defect boundaries. A significant portion of the evaluation focused on the generalization capabilities of the models. DeepLabV3+ demonstrated remarkable adaptability, not only performing well on the diverse images of the WALL-E dataset but also showing robust generalization on arbitrary internet images and out-of- distribution samples. This indicates the model’s ability to extend beyond its training parameters and adapt to new, unseen scenarios, which is essential for practical applications where variable conditions are common. Further assessments were made on images without defects to evaluate the model’s precision in iden- tifying defect-free areas, thereby reducing the likelihood of false positives. This aspect of the model’s performance is critical in practical scenarios, where the ability to distinguish between damaged and intact structures can significantly impact maintenance decisions and safety assessments. This thesis highlights the significant potential of using semantic segmentation in the realm of struc- tural health monitoring. The advanced capabilities of DeepLabV3+, coupled with the comprehensive and diverse WALL-E dataset, set a new benchmark in the field, providing a strong foundation for future research and practical applications in facade inspection. This work not only advances the technology in automated building inspections but also opens up new avenues for further enhancements and applica- tions in urban infrastructure maintenance.

Abstract

Federated learning offers a promising approach to learn collaboratively across multiple devices or agents. This collaboration allows leveraging a vast amount of data while preserving privacy by keeping the raw data local. However, this approach is not void of challenges. This thesis explores some of these challenges in the context of multi-armed bandit (MAB) problems. MAB problems are a type of decision-making problem where an agent repeatedly chooses between multiple actions, each with an unknown reward, and aims to maximize the total reward over time. Federated MABs introduce unique fairness concerns. For instances, for a fair algorithm, the learning process should not favor certain actions over others to an extreme degree, especially if those actions are beneficial to specific sub-populations. In addition to fairness, privacy is another key challenge. While federated learning avoids sharing raw data, the learning process itself can leak information about individual data points. This thesis proposes two novel algorithms (and variants) to address these challenges. The first algorithm, P-FCB, tackles a specific type of MAB problem with constraints in a federated setting. It promotes fairness by optimizing for the collective benefit of all users. The second algorithm, Fed-FairX-LinUCB, focuses on achieving fairness and privacy guarantees in a more complex scenario where actions have additional context. It ensures that action selection is fair and avoids situations where some actions are never chosen. Additionally, both these algorithms provide differential privacy guarantees, a formal guarantee that ensures the learning process reveals minimal information about any individual data point. By exploring these algorithms, this thesis aims to contribute to the development of federated learning for MAB problems that is both fair and privacy-preserving.

Abstract

Sleep plays a critical role in maintaining overall health and well-being, impacting both psychological and physiological functions. However, sleep disorders such as insomnia, obstructive sleep apnea (OSA), and narcolepsy are becoming increasingly prevalent, significantly affecting individual’s quality of life. Effective sleep stage classification is essential for diagnosing and treating these disorders. Traditional methods like polysomnography (PSG) are the gold standard for diagnosing sleep disorders, but they are expensive, cumbersome, and not easily accessible due to the need for specialized equipment and trained personnel. Moreover, PSG relies heavily on electroencephalogram (EEG) data, which, despite its effectiveness, presents practical challenges in non-clinical settings due to the complexity of electrode placement and patient discomfort. Innovative strategies in wearable technology using supervised deep learning techniques offer a promising alternative for enhancing sleep stage classification, making it more accessible and user-friendly. This work aims to overcome these limitations by exploring the potential of using electrooculogram (EOG) signals for sleep stage classification. EOG signals, which are non-invasive and can be more easily integrated into wearable technology, offer a promising alternative for long-term sleep monitoring and provide a more comfortable experience for patients. The core contribution of this research is the design and implementation of a novel SE-ResNet-Transformer model. This model architecture combines the strengths of Squeeze-and-Excitation Residual Networks (SE-ResNet) and Transformer encoders, tailored to handle the unique characteristics of EOG signals. The SE-ResNet component enhances feature extraction by adaptively recalibrating feature responses, allowing the model to focus on the most informative aspects of the EOG signal. The Transformer encoder component captures the temporal dependencies within the EOG signals, enabling the model to understand the dynamic transitions between sleep stages. The performance of the SE-ResNet-Transformer model was rigorously evaluated using three publicly available datasets—SleepEDF-20, SleepEDF-78, and SHHS. The model demonstrated superior accuracy and robustness across different datasets, with its EOG-based performance comparable to models trained with EEG data. This suggests that EOG is a viable alternative for sleep monitoring, particularly relevant for the development of wearable technology. Extensive ablation studies were conducted to fine-tune the model’s parameters and configurations, providing valuable insights into optimal settings for window sizes and strides, and enhancing the model’s efficiency and performance. Techniques like 1D-Gradient-weighted Class Activation Mapping (Grad-CAM) and t-distributed Stochastic Neighbor Embedding (t-SNE) plots were employed to visualize the model’s decision-making process, emphasizing transparency and interpretability. In addition, this work introduces the pioneering Indian SLEEP dataset of ischemic Stroke patients (iSLEEPS) which addresses a gap in the availability of India-specific sleep disorder data, with a particular focus on the impact of ischemic stroke on sleep health. This dataset is comprehensive, including detailed clinical annotations that cover patient demographics, diagnoses, and sleep stages. Cutting-edge deep learning models, including CNN-based, LSTM-based, and Transformer-based architectures, were trained on raw single-channel EEG and EOG signals. Among these, the Transformer-based model showed remarkable performance in sleep stage classificatzions using single-channel EOG data. The successful validation of these models on the iSLEEPS highlights their potential to enhance diagnostic tools and therapeutic strategies, thereby contributing to improved patient outcomes and fostering global collaboration in sleep research. In conclusion, this thesis presents advancements in the field of automated sleep stage classification using wearable technology and supervised deep learning. By leveraging EOG signals and deep learning architectures, this research offers promising solutions for more accessible, accurate, and comfortable sleep monitoring, ultimately aiming to bridge the gap between clinicalgrade sleep monitoring and everyday health management tools, thus enhancing sleep health and overall wellness