Abstract

The critical brain hypothesis suggests that neural networks self-organise into a reverberating regime poised between order and disorder. In this state, the system acquires emergent properties such as scale invariance and long-range correlations, which are thought to provide an optimal balance for information processing, transmission, and storage. While deviations from this state have been implicated in neurological conditions such as Alzheimer’s Disease (AD) and Autism Spectrum Disorder (ASD), the precise nature of this phase transition remains an open question. This thesis investigates the critical dynamics of the human brain using resting-state functional magnetic resonance imaging (fMRI) data. It explores frameworks beyond standard ferromagnetic models to consider the complex and frustrated dynamics characteristic of spin-glass systems. We first employed the Pairwise Maximum Entropy Model (pMEM) to infer the functional connectivity of the brain and simulate its thermodynamic behaviour. By analysing the paramagnetic-ferromagnetic phase transition, we determined the critical temperature (Tc) for each subject based on peak susceptibility. While we observed a trend where AD subjects exhibited a slightly higher Tc than healthy controls, which suggests a shift toward ordered dynamics, this distinction was weak and not statistically significant. Similarly, static order parameters derived from a Sherrington-Kirkpatrick model, including uniform and spin-glass susceptibility, did not yield robust differentiations between the groups. These observations suggest that static equilibrium measures and simple ferromagnetic transition models may be insufficient to capture the specific pathological dynamics associated with neurodegeneration. To address this limitation, we extended our investigation to the temporal domain. We hypothesized that the brain operates in the vicinity of a spin-glass phase characterized by a rugged energy landscape. By analysing the temporal autocorrelation of BOLD signals, we observed that neural relaxation tends to follow a stretched exponential form (β < 1) rather than a simple exponential decay. This pattern is a hallmark of disordered systems where frustrated interactions create a hierarchy of metastable states. Unlike simple ferromagnetic phases that tend toward global uniformity, the metastability inherent in the spin-glass phase allows the system to freeze information in local configurations without locking the entire network. This property is theoretically essential for supporting long-term memory and flexible cognitive processing, features that simple order-disorder transitions cannot adequately explain. The biological substrate for this metastability in the brain comes from the electro-chemical balance of excitatory and inhibitory (E/I) synaptic interactions. Metastable states and the subsequent ability to freeze in local configurations are theoretically essential for supporting long-term memory and flexible cognitive processing, features that simple order-disorder transitions cannot adequately explain. Furthermore, our analysis indicated a significant distinction in the characteristic relaxation time (τ ). While healthy brains exhibited longer relaxation times consistent with “critical slowing down,” AD brains displayed reduced relaxation times. This reduction implies a loss of temporal integration where the neural networks in AD recover too quickly from spontaneous fluctuations, failing to sustain the reverberating activity that might be required for long-term memory. These findings support the hypothesis that brain criticality is of the spin-glass type. They also suggest that temporal relaxation dynamics offer a sensitive, model-independent biomarker for detecting the loss of cognitive integration in neurodegenerative disorders.

Abstract

The agriculture sector’s contribution to the GDP is expected to be around 18.2% in 2025. While this percentage has been declining over time, agriculture remains a vital sector, particularly for employment and food security. The Indian agricultural sector faces challenges such as low crop productivity, inadequate infrastructure, and increasing vulnerability to climate change. Among multiple factors, unscientific agricultural practices by farmers are one of the factors causing low crop productivity. In spite of advances in ICTs, Data Science and AI, Indian farmers are facing issues in getting timely, actionable scientific agro advisories. Farmers experience difficulty obtaining real-time agricultural advice from call centers and web portals, while radio, SMS, and voice-based services deliver only generic information. Farm-specific advisory systems like eSagu suffer from scalability issues. The Plantix kind of image-based systems are probabilistic and require high-quality, huge training data. At IIIT Hyderabad, an effort has been made to develop a question-answer-based smartphone-based application (or app), called Crop Darpan. The basic idea is as follows: crop diseases can be identified by confirming the presence/absence of visual symptoms by farmers. There is an opportunity to build a system to help farmers by organizing visual symptoms in a hierarchical structure by forming a parent-child relationship among the symptoms. A prototype for the Cotton crop was developed during 2017-22 under the Indo-Japan project by collaborating with Telangana Agricultural University. The response from the farmers, agricultural scientists, and other stakeholders was encouraging. The framework employs knowledge acquisition protocols, hierarchical knowledge bases, and dynamic question popping models. While trying to develop the Crop Darpan app for the rice crop, we have identified several issues with the existing Crop Darpan framework for developing the cotton app. A straightforward extension, i.e., extending the framework employed for the development of the Cotton crop app to develop the Rice Crop app, was found to be inadequate, as the process of developing the Rice crop app presented unique challenges, such as problems that appear similar in both the seedling and post-transplantation stages, and also require different advisories. In this thesis, we have extended (i) the scalability and usability of the Crop Darpan framework by adding extensions and (ii) developed it for the rice crop. The following enhancements to the Crop Darpan framework were made to enhance scalability and usability: (i) a flexible database schema to support future expansion to new crops and languages (ii) a direct advice option for known problems (iii) dynamic diagnostic algorithms using utility-based questioning and regional relevance (iv) multimediaassisted interfaces to clarify symptoms and (v) a comprehensively improved user interface for all stakeholders. We have also presented how the enhanced framework is being used to develop the Crop Darpan vi vii app for the Rice crop, and reported the evaluation results. The field evaluation results from the stakeholders are encouraging. Overall, we present a scalable Crop Darpan prototype, which can be extended to other crops and languages.

Abstract

Secunderabad, once a city in princely Hyderabad State in colonial South Asia, was shaped critically by the arrival of immigrant communities from different parts of the subcontinent. One such group was the educated, professional, Tamil-speaking immigrants from the Madras Presidency. In this thesis I investigate the historical presence of immigrants of Tamil ancestry in Secunderabad and analyse how they imagined and maintained a sense of collective identity and memory. The immigrants of Tamil ancestry in Secunderabad are a micro-minority considering their lack of representation in the archives. The keepers of the history of the Tamil-heritage immigrants of Secunderabad are the individuals and the various associations and institutions they are a part of, through which they form a sense of community and shared identity. Where official records falter, oral history, cultural and commemorative interventions, community initiatives, and works of fiction are what keeps the history alive. Through my investigation of the history of immigrants of Tamil ancestry in Secunderabad I demonstrate the need to look beyond the standard paradigms of social research to study such a group. In this thesis I examine private archives, commemorative souvenirs, community records, and works of fiction in conjunction and contestation with official records and argue that the people of Tamil heritage in Secunderabad have maintained a tenuous, somewhat shifting notion of “a Tamil community

Abstract

Music listening is a pervasive human behaviour that supports emotion regulation, social connectedness, and identity formation. However, not all engagement with music is adaptive. Maladaptive music engagement, characterised by using music for rumination, avoidant coping, and mood worsening, have been linked to depression risk. Yet the underlying individual differences shaping such behaviours have rarely been looked into. This thesis therefore examines how dispositional and affective traits interact with environmental context to drive maladaptive music engagement through two complementary studies. Study 1 (N = 318) investigated whether maladaptive music use could be modelled by personality and affective traits, as well as by emotional contagion and types of musical reward. Maladaptive music engagement was found to be positively associated with Neuroticism, Personal Distress, and contagion of negatively valenced emotions, such as Sadness and Anger. Structural equation modeling (SEM) highlighted the importance of Personal Distress in mediating maladaptive music use. Study 2 (N = 680) expanded the investigation by examining the role of Sensory Processing Sensitivity (SPS), a heritable, neurobiological trait, as a driver of maladaptive music engagement. Since context plays a crucial role in determining adaptive or maladaptive music use, this study was conducted in India rather than the typical W.E.I.R.D. (Western, Educated, Industrialised, Rich, and Democratic) settings. Using exploratory factor analysis, principal component analysis, regularised partial correlation networks, and SEM, results showed that sensitivity to external demands elevated psychological distress, which in turn increases maladaptive music use. Understanding these differences in environmental contexts may provide us with more precise methods for designing highly personalised music based interventions and tools

Abstract

This thesis details the design, development, and implementation of Śabdabrahman Exercise Platform (SBE). SBE is an interactive online system for learning Sanskrit through a linguistic approach. Motivated by the challenges students face in mastering multiple linguistic aspects—script, phonetics, morphology, syntax, and semantics—and the difficulty for instructors to provide timely, individualized feedback, SBE offers a comprehensive solution. Based on the textbook ”Śabdabrahman: a linguistic introduction to Sanskrit” by Scharf (2022a), the platform guides students through exercises in transliteration, sandhi and sandhi analysis, morphological identification, syntax analysis, translation and more. Key contributions include the development of diverse interactive exercise types, a system for immediate and focused feedback, user progress tracking, and an instructor interface for monitoring student performance. The platform features a content management system for efficient handling of Sanskrit learning materials encoded in the SLP1 scheme, ensuring accurate representation and processing of text. SBE employs a multi-tiered architecture with a React JS frontend, a Python Flask backend, and a SQLite database. This work addresses the scarcity of specialized digital tools for classical language education by providing a robust platform that facilitates structured learning and personalized feedback, thereby enhancing the Sanskrit learning experience.

Abstract

Machine learning enabled systems (MLS) have become increasingly prevalent in modern softwareintensive applications. These systems are typically deployed for long periods and operate in environments where data, workloads, and operating conditions continuously evolve, leading to various runtime uncertainties such as data drift and model degradation. To remain useful in such settings, MLS must operate sustainably over the long run. Sustainability in this context is multi-faceted and extends beyond short-term performance. It encompasses technical, environmental, economic, and social dimensions whose effects accumulate during prolonged operation. While many engineering practices focus on improving individual aspects of system performance, long-term sustainability requires balancing these dimensions under persistent uncertainties. Machine Learning Operations (MLOps) has emerged as an engineering discipline to support the deployment and maintenance of machine learning models in production. By automating data processing, model training, deployment, monitoring, and retraining, MLOps pipelines significantly improve technical sustainability. However, common MLOps practices such as frequent or periodic retraining often introduce substantial computational overhead. As a result, other sustainability dimensions, particularly environmental and economic sustainability, are adversely affected. This thesis investigates how self-adaptation can be applied as an architectural mechanism to manage sustainability concerns in MLOps pipelines across multiple dimensions. It proposes an approach that combines design-time sustainability goal classification using Decision Maps with runtime selfadaptation regulated through adaptation boundaries. These concepts are realized within a self-adaptive architectural framework based on the MAPE-K loop, enabling deliberate and controlled adaptation under runtime uncertainty. The approach is grounded and evaluated using a Digital Twin for an Intelligent Transportation System, where traffic flow prediction serves as a representative long-running and data-intensive use case. The evaluation shows that regulated self-adaptation enables MLOps pipelines to balance predictive performance and resource consumption more effectively than static and periodically adaptive baselines, while reducing unnecessary retraining and limiting energy usage. Building on this framework, the thesis also introduces a reusable self-adaptive exemplar that operationalizes sustainability-aware adaptation in MLOps pipelines and supports systematic experimentation and reuse. Overall, this thesis demonstrates that treating sustainability as an explicit architectural concern and regulating adaptation decisions at runtime provides a principled foundation for the long-term operation of machine learning enabled systems in dynamic environment

Abstract

Road safety remains a critical global issue, particularly in rapidly developing regions like India, which reports over 461,000 road traffic accidents annually. This thesis explores the integration of AIdriven Advanced Driver Assistance Systems (ADAS) and Collision Avoidance Systems (CAS) in public transport to enhance road safety. Increasing urbanization and mixed traffic conditions in Indian cities make such interventions vital for reducing accidents and improving driver behavior.This research investigates how AI-enabled safety systems influence driver behavior, infrastructure planning, and policy frameworks. Using a mixed-methods approach—including field observations, driver interviews, and policy analysis—it assesses the effectiveness of AI interventions, driver adaptation, and socio-technical challenges, and offers policy recommendations. It also examines AI’s potential role in proactive accident prevention. Findings indicate that ADAS significantly improves safety by enhancing driver awareness and response time through real-time alerts and behavioral feedback. However, effective deployment depends on addressing human factors such as training, perceptions of surveillance, and infrastructural readiness. The study also notes concerns about technology acceptance, particularly job security and driver autonomy. The thesis further examines how ADAS can inform policy and urban transport planning. AI-generated insights help identify high-risk zones, optimize infrastructure, and support targeted safety programs. Nonetheless, barriers such as low technological literacy and resistance to behavioral change must be addressed to realize AI’s potential fully. A human-centered AI approach is essential, ensuring technology augments rather than replaces human judgment. This includes incorporating user feedback, adaptive learning, and socio-cultural factors into design and deployment. Engagement with drivers, policymakers, and transport authorities through continuous training is key to fostering acceptance and long-term sustainability. By integrating technological and anthropological perspectives, this thesis contributes to the growing literature on AI and road safety. It offers actionable insights for policymakers, transport planners, and developers, advocating a data-driven yet human-centric approach to safer mobility in India and beyond. It also highlights the ethical and governance challenges of AI, underscoring the need for inclusive policy and accountability frameworks. Ultimately, this study proposes a roadmap for integrating AI into road safety efforts, balancing innovation with human needs to support an equitable and sustainable transportation ecosystem.

Abstract

Whole slide images (WSIs) contain rich information for computational pathology, yet systematic evaluations of cross-organ and multi-cohort generalization remain limited. This thesis addresses these challenges through two complementary studies. In the first study, patch-level convolutional neural networks (CNNs) were trained on 9,792 slides from The Cancer Genome Atlas (TCGA), spanning 11 cancer subtypes across seven organs, to distinguish cancerous from normal tissue. Both within-organ and cross-organ inference were evaluated, revealing that cancers such as breast, colorectal, and liver can be reliably detected by models trained on other organs. Strong transferability was also observed between subtypes within an organ, such as kidney and lung. To investigate these patterns, feature similarity, overlap in high-attention regions, and nuclear geometry were analyzed, all of which showed positive correlations with cross-organ transfer performance. The second study focuses on lung cancer in the Indian population through the introduction of IPD-Lung, a curated dataset of adenocarcinoma and squamous carcinoma cases. Benchmark evaluations were established using multiple instance learning (MIL) models, with additional experiments exploring model transferability from publicly available TCGA-Lung dataset. Stain normalization methods, particularly Macenko, reduced domain discrepancies to some extent but did not fully bridge performance gaps, especially for squamous carcinoma. A domain adaptation strategy based on a gradient reversal layer (GRL) similarly yielded limited improvements. In contrast, models trained directly on IPD-Lung achieved substantial performance gains, with further enhancements obtained by training on expert-annotated regions of interest (RoIs). Together, these studies demonstrate that deep learning models can uncover meaningful cross-organ similarities in cancer histopathology. At the same time, they highlight the difficulty of generalizing across cohorts, even for the same cancer subtype, underscoring the influence of dataset-specific factors such as scanners and staining protocols. Finally, the findings show that expert annotations can substantially improve performance in small and imbalanced datasets, providing a practical pathway to more robust and clinically relevant computational pathology models

Abstract

Mangoes hold immense economic and cultural significance in India and globally, making accurate yield prediction essential for optimizing domestic consumption, enhancing international trade, and supporting farmer decision-making. Traditional estimation methods, such as manual counting, are laborintensive, error-prone, and impractical for large-scale orchards, highlighting the need for automated solutions. In this work, we present a novel time-series vision dataset for mango yield prediction, capturing high-resolution images of 32 trees across eight spatial orientations over two growing seasons to analyze fruit growth and flowering dynamics. Weather data from the nearest weather station was also collected. To reduce manual effort, we developed a semi-automatic annotation pipeline by integrating YOLO for object detection and SAM for precise segmentation. Benchmarking experiments were conducted for both segmentation and detection tasks. Building on these results, we propose a baseline framework for early-stage yield prediction using flowering cues, combining SegFormer-based segmentation with a regression pipeline and incorporating contextual features such as weather and scale. Our multimodal analysis shows that weather factors complement visual features, and their integration further improves prediction performance. The single-image model, based on the SegFormer-B1 encoder, achieved a mean absolute error (MAE) of 7.21, an R2 of 0.80, and a mean squared error (MSE) of 96.83. These results highlight the potential of vision-based models for yield estimation from early-stage flowering cues. To the best of our knowledge, this is the first work to address mango yield prediction using images from the flowering stage in combination with weather data.

Abstract

Groundwater forms a vital component of freshwater resources, particularly in developing countries like India, where borewells are extensively used for agricultural and domestic water needs. However, unsustainable extraction and limited monitoring have led to significant depletion of groundwater levels, exacerbating water scarcity. Traditional methods for measuring borewell water levels are often manual, error-prone, and lack temporal resolution. This thesis presents the design and deployment of an IoT-enabled borewell monitoring system based on a novel string-tension mechanism that automates the measurement of water depth with minimal human intervention, while providing real-time, automated borewell monitoring. The proposed system leverages low-cost sensors and microcontrollers to periodically record water levels, which are transmitted to cloud platforms for storage and analysis. The device is designed to accommodate the geometrical constraints of borewells and withstand practical deployment challenges, including those involving submersible pump infrastructure. Multiple iterations of sensor integration, including FSRs, encoders, and limit switches, are discussed, with emphasis on reliability, calibration, and mechanical design. To ensure data quality, the raw measurements undergo a multi-stage preprocessing pipeline that includes outlier removal, median filtering, and dynamic window-based smoothing. Realworld deployments demonstrate the system’s ability to produce reliable, high-resolution timeseries data across different locations and elevations. The collected data not only enables real-time water level monitoring but also supports longterm groundwater studies. By enabling scalable and cost-effective data collection, the system facilitates greater awareness of groundwater dynamics, promotes rainwater harvesting practices, and contributes empirical inputs for data-driven policymaking and water resource governance.