Abstract

Document question answering has witnessed substantial progress with the development of large visionlanguage models. Despite these advances, most existing systems remain limited in their interpretability,
as they typically generate answers without explicitly and precisely identifying the visual evidence on
which those answers are based. This deficiency is particularly pronounced in real-world documents,
where supporting evidence may appear in curved text, be fragmented across multiple spatially separated
regions, be embedded within tables or charts, or exist at different semantic levels such as phrases, lines,
and blocks. Under such conditions, conventional grounding approaches based on coarse bounding boxes
are often insufficient, as rectangular regions fail to capture the irregular and fine-grained structure of
document evidence.
This thesis addresses these limitations through M3Grounder, a novel mask-based framework for
grounded document question answering. In contrast to prior approaches that represent evidence using
rectangular coordinates, the proposed framework formulates grounding as a segmentation problem
and predicts pixel-level evidence masks that are tightly aligned with the answer generation process.
The model autoregressively generates answer text interleaved with special grounding tokens, enabling
explicit correspondence between individual answer spans and their associated supporting regions. This
formulation naturally supports multi-span grounding, in which a single answer may depend on multiple
disjoint regions, and multi-granular grounding, in which evidence may be localized at the phrase, line,
or block level depending on the semantic scope required by the question. This design enables answer
generation interleaved with grounding tokens and hierarchical evidence prediction.
To realize this formulation, the thesis introduces a set of architectural and training mechanisms
designed to improve the precision and faithfulness of document grounding. In particular, M3Grounder
integrates document understanding with segmentation-based evidence prediction, employs a bleedsuppression objective to reduce spillover into irrelevant neighboring regions, and incorporates a hierarchical enclosure mechanism that encourages spatial consistency across phrase-, line-, and block-level masks.
By enforcing containment relationships across these granularities, the framework more faithfully reflects
the hierarchical structure of document layouts and supports more interpretable evidence attribution. Collectively, these components enable finer and more reliable localization than prior box-based grounding
methods, especially in documents containing irregular text geometry or densely structured layouts.
Beyond model design, this thesis also introduces GroundingDocQA, a large-scale data generation
pipeline for grounded document question answering, and GroundingDocQA-Bench, a benchmark for
the joint evaluation of answer correctness and grounding fidelity. The dataset construction pipeline is designed to support a broad range of document types, including text-rich pages, curved-text documents,
and graphics-heavy documents such as charts and infographics. This enables supervision for a wide
variety of grounded reasoning scenarios that extend beyond conventional extractive settings. The
benchmark complements standard answer-centric evaluation by explicitly assessing whether predicted
evidence is spatially accurate, semantically appropriate, and faithful to the generated answer. In this way,
the thesis promotes a more comprehensive evaluation paradigm for document question answering, one
that treats evidence quality as a first-class criterion alongside answer accuracy.
Extensive empirical results demonstrate that mask-based grounding constitutes a more expressive
and effective alternative to coarse region attribution for document question answering. The proposed
framework yields improvements in evidence faithfulness, better captures irregular and spatially distributed
support, and provides stronger grounding performance across challenging document layouts. More
broadly, this thesis argues that the next generation of document question answering systems should not be
evaluated solely by whether they answer correctly, but also by whether they can provide precise, verifiable,
and semantically appropriate evidence for their predictions. In this respect, the thesis advances document
question answering toward more interpretable, trustworthy, and practically deployable system

Abstract

The current project deals with the digitalization of the ICMR-NCTB toolkit through the use of digital
platforms while preserving the diagnostic validity of traditional pen-and-paper tools. It reports on the
design and development of a digital adaptation of verbal components within the ICMR-NCTB cognitive
assessment toolkit. The study aimed at translating the original test structure, its administration procedures, and scoring methods into an interactive, device-based format while preserving the functionality
of the traditional pen-paper tools. A prototype was developed that administered selected verbal tasks,
recorded participant responses by audio capture, and stored results in a structured digital format (JSON)
for subsequent and continuous evaluation. Preliminary validation was conducted with a cohort of young,
academically oriented students who were administered the selected verbal tasks in both digital and traditional modes. Results indicated that the digital version can replicate core assessment functions, although
more extended validation efforts are required to establish generalizability across diverse age groups and
clinical populations. Challenges identified include automated audio processing, user experience design,
and data-handling infrastructure, representing key avenues for further refinement. The contribution provides the basic framework for a fully digitized, scalable, clinically reliable tool for cognitive assessment,
matching current technological standards.

Abstract

Index Terms: Falsetto register, Glottal activity, High arousal speech, Intrinsic variations, Lombard speech, Modal register, Single frequency filtering, Speech synthesis, zero time windowing, zero frequency filtering.
High-arousal speech is typically produced in emotionally charged situations, during longdistance communication, or in the presence of background noise. At the extreme, this type of
speech is called "shouted speech." Emotions such as anger and happiness also commonly result
in high-arousal vocal expressions. This thesis analyzes, detects, and synthesizes high-arousal
(loud) speech.
The first part of the thesis investigates acoustic and glottal features critical for distinguishing high-arousal speech from neutral speech. Prior studies on high-arousal speech have utilized
segmental features such as spectral tilt and energy ratios across low and high-frequency bands.
Others have investigated subsegmental features derived from glottal flow waveforms, which are
often difficult to obtain. In contrast, this thesis adopts a suprasegmental approach. The glottal
activity is parameterized using the open quotient Oq, estimated from electroglottograph (EGG)
signals. In addition, an acoustic feature--the ratio of high to low-frequency spectral energy
(β) is derived using the zero-time windowing (ZTW) method, which provides high temporal resolution for spectral analysis. High-arousal speech is further analyzed across different
loudness levels using intrinsic variations among six sound units. Key acoustic parameters include the fundamental frequency (F0), the abruptness of excitation (Im) and short-term energy
(Se). These are computed around the glottal closure instants, and are derived using the zerofrequency filtering (ZFF) method. Building on this analysis, a novel approach is proposed for
high-arousal speech detection that does not require a reference to neutral speech, unlike most
machine learning-based methods. Instead, intrinsic suprasegmental variations in speech are
leveraged, offering a reference-independent and computationally efficient detection method.
Further, this thesis focuses on Lombard speech generation using different approaches. For
synthesis, both analysis-by-synthesis and data-driven methods are explored. Some studies use
spectral and source features to generate high-arousal speech, but their relative impact on synthesis is unknown. Hence, we study the fundamental frequency along with the duration modification of Lombard's speech. At the source level, previous studies have shown an increase in
fundamental frequency (F0), but the importance of F0 contour shape and F0 contour level in
Lombard speech is unclear. These perceptual studies carried out in this thesis highlight the role
of F0 contour shape and F0 contour level along with the durations. Six different experiments
were conducted by modifying durations and both F0 contour shape and level. These experiments are performed on samples from the Hurricane Challenge-2013 database. Perceptual
results show that the contribution of the F0 contour level is more than the F0 contour shape and
duration.
Finally, Lombard speech synthesis is achieved via transfer learning. Transfer learning eliminates the need for huge amounts of data to synthesize Lombard speech. This thesis uses
different neural parameter generation and vocoder models and compares the ability to adapt
to minimal Lombard speech that is available. In addition, a study based on metric learning
is conducted. Metric learning was initially studied on natural speech to enhance speech intelligibility in a noisy environment. The approach is adapted to the TTS system to improve
speech in noise. Metric learning is used to learn an intelligibility metric that generates a speech
with unique characteristics, such as increased loudness and higher pitch, compared to normal
speech. By aligning the learned metric with the acoustic features of the target speech domain,
the model improves its ability to synthesize realistic and intelligible Lombard speech from
textual input.

Abstract

Reliable and generalizable motion planning is essential for deploying robotic manipulators in realworld environments, where uncertainty, variability, and previously unseen conditions are common.
Classical optimization-based motion planners can be applied to any new scene without requiring taskspecific training, offering strong adaptability; however, they rely heavily on manually designed cost
functions tailored to each scene, which limits their scalability and performance in complex settings . In
contrast, deep-learning-based motion planning methods achieve high success rates across diverse trained
scenarios, but they often struggle to generalize to out-of-distribution scenes that differ from training data.
This thesis presents two complementary contributions that enhance generalization in robotic motion
planning and policy initialization for unseen in-distribution tasks, with extensions to multi-task learning.
First, we introduce EDMP (Ensemble-of-costs-guided Diffusion for Motion Planning) [7], detailed
in Chapter 3 which bridges classical planning with learned generative models. EDMP is trained on
diverse, kinematically valid trajectories and, at inference time, incorporates scene-specific cost signals,
such as collision and reachability costs, to guide the diffusion process. By leveraging an ensemble of
costs rather than a single handcrafted objective, EDMP produces feasible, diverse trajectories while
preserving the adaptability of classical planners. Our method exceeds the performance of classical
motion planners and matches the performance of state-of-the-art learned planning methods while offering
strong generalization to novel scenes. This work has been accepted for publication at IEEE International
Conference on Robotics and Automation (ICRA) 2024.
Second, we propose GenHyper, detailed in Chapter 4, a framework focused on zero-shot transfer
within an in-distribution task family. Using adversarially trained hypernetworks, GenHyper generates
strong policy initializations that perform well on unseen in-distribution tasks (demonstrated on MuJoCo
continuous-control benchmarks) with little or no environment interaction. Importantly, we also extend
this zero-shot framework to multi-task robotic manipulation settings using residual learning in latent
policy parameter space. Across benchmarks, GenHyper outperforms existing in-context, meta-RL, and
hypernetwork baselines on zero-shot transfer and enables faster adaptation with minimal interactions.
This work was accepted at the AAAI 2025 GenPlan and PRL Workshops.

Abstract

This doctoral research presents a comprehensive and multi-faceted investigation into automated table
understanding, arguing that robust and reliable solutions demand an approach that evolves from foundational structural parsing to address the pressing real-world requirements of data privacy and auditable
reasoning. Tables are information-rich, structured objects that serve as a cornerstone for conveying
complex data, yet their automated parsing is a formidable, long-standing challenge in document intelligence. The core of this challenge lies in Table Structure Recognition (TSR), the process of transforming
a table image into a structured, machine-readable format. The difficulty is rooted in the immense visual diversity of tables, with complexities such as spanning cells, multi-line text, and the absence of
ruling lines often causing traditional and early deep learning methods to fail. This body of work charts
a clear research trajectory that begins with the development of a novel framework for TSR, TabStructNet, and progressively refines this methodology to handle increasing visual complexity. The research
then pivots to address critical non-functional requirements, pioneering TabGuard, a novel framework for
privacy-preserving TSR, and finally extends its scope from structure to trusted reasoning by introducing EviFiVQA, a benchmark for financial Visual Question Answering (VQA) that establishes evidence
localization as a core tenet of auditable AI. This journey from pixels to privacy and proof marks a significant contribution to the field.
The core methodology advanced throughout this research is anchored in a powerful two-step paradigm
that mirrors human cognitive processes: a top-down decomposition followed by a bottom-up reconstruction. In the top-down phase, the table image is decomposed into its fundamental constituent parts--the
individual table cells--through an object detection model. In the bottom-up phase, the global table
structure is reconstructed by learning the spatial and logical associations between the detected cells. A
cornerstone of this research is the novel insight that TSR performance can be dramatically improved by
encoding human intuition about table structure directly into the learning objective. This was achieved
through a series of innovative, cognitive-inspired loss functions that act as structural regularizers, including an Alignment Loss to enforce a grid-like structure, a Continuity Loss to ensure adjacent cell
boundaries are contiguous, and an Overlapping Loss to penalize spatial conflicts. This approach is
marked by a clear architectural evolution, beginning with TabStruct-Net, which combined a modified
Mask R-CNN with a Dynamic Graph Convolutional Neural Network, and culminating in TabStructNet V2, which introduced a Hierarchical Local-Attention Vision Transformer (HLVIT) backbone and a
highly efficient self-attention layer to achieve state-of-the-art performance and scalability.Building upon the foundation of accurate structural recognition, the research pivots to address a critical
real-world barrier to the adoption of document AI: data privacy. The research introduces TabGuard, a
pioneering framework that enables privacy-preserving TSR through a novel client-server architecture.
In this model, the client performs content masking locally, blacking out all text regions before sending
only the anonymized image to the server for structure recognition. This effectively decouples structural
understanding from content recognition, ensuring sensitive data is never exposed. The key technical
innovation that makes this possible is the Table Grid Approximator (TGA), a server-side module that
analyzes the spatial distribution of masked text contours to infer a coarse structural grid. This grid serves
as a powerful inductive bias, allowing for the dynamic generation of high-quality, layout-specific anchor boxes that enable the detection network to converge accurately even without access to text content.
The success of TabGuard provides compelling evidence that a table's structure can be robustly inferred
purely from its spatial layout cues, independent of its content.
The final thrust of this research extends beyond structural parsing to the higher-level goal of building
trustworthy AI systems for semantic document understanding, explored in the high-stakes domain of
financial analysis. Identifying a critical gap in existing VQA benchmarks, which lack support for the
complex numerical and hierarchical reasoning required for financial documents, the EviFiVQA benchmark was created. Its most significant contribution is the requirement of evidence localization: for
each question, a model must produce not only the correct answer but also the bounding boxes of all
table cells used to compute it. This feature transforms VQA from a black-box task into a transparent
and interpretable one, providing a direct mechanism for auditability by allowing an ana

Abstract

Inspired by the natural brachiation of primates, this thesis presents a learning-based control
framework for generating continuous swing-grasp locomotion in a robotic system equipped
with high-degree-of-freedom anthropomorphic hands. Unlike conventional brachiation robots
that rely on hooks or low-DOF grippers, the proposed system integrates a two-link acrobot with
dual 16-DOF anthropomorphic hands, enabling dynamic grasping and release under highly
nonlinear swing dynamics.
The control framework combines model-based trajectory optimization with data-driven learning to address the limitations of purely model-based controllers in high-DOF, contact-rich settings. Energy-efficient swing trajectories are first generated using trajectory optimization on a
simplified acrobot model. These trajectories provide expert demonstrations for learning-based
controllers and reference motions for execution on a full 35-DOF MuJoCo simulation. While
reference tracking achieves accurate performance under nominal conditions, it degrades under perturbations and model mismatch, motivating the use of imitation learning methods for
improved adaptability.
Several learning-based controllers are implemented and evaluated, including Behavior Cloning
and its recurrent extension, Dataset Aggregation, Generative Adversarial Imitation Learning(GAIL), Adversarial Inverse Reinforcement Learning(AIRL), and a residual diffusion policy for torque refinement. These methods learn from optimized demonstrations and enable recovery from off-trajectory states, improved robustness to dynamic perturbations, and smoother
torque profiles during swing execution. Grasping is coordinated through a proximity-triggered
controller that synchronizes high-DOF finger motion with global swing dynamics.
Comprehensive evaluation in MuJoCo demonstrates clear trade-offs among accuracy, energy consumption, timing, and robustness. Among all learning-based approaches, AIRL achieves
the most reliable and energy-consistent brachiation. The results highlight the effectiveness of
adversarial and generative imitation-learning methods for dynamic, high-DOF brachiation and
point toward their applicability in real-world scenarios such as power-line inspection, forest
canopy traversal, and search-and-rescue operations.

Abstract

Mathematical modeling is an essential exercise for the scientific community, for understanding the
processes at play, and for predicting how they evolve over time. In the context of modern epidemiology,
it becomes necessary in understanding and controlling the spread of infectious diseases. Along with
public health interventions being logistically challenging, their success is deeply affected by human
behavior, which can either improve or negatively affect their efficacy. Thus, this thesis attempts to study
two key aspects of real-world epidemics.
This thesis first discusses the role of population mobility and interventions (in the form of testkits) in
a meta-population context. In the peer-reviewed study we published in Chaos, we study the impact of
population diffusion on a network of communities where an intervention in the form of testkits is nonuniformly distributed. Our key finding is the emergence of synchronization patterns, where communities
cluster based on their access to the intervention, a phenomenon dependent on the rate of migration
between the communities.
The second factor we investigate is the impact of opinion dynamics on epidemic evolution. Since
humans are social beings, they are affected by the opinions of their peers, while also accounting for the
perceived risk brought about by the severity and scale of the disease. We generalized a coupled opinionepidemic model, moving one step above the all-to-all network assumption to a model that generalizes to
arbitrary network topologies, especially Barabasi-Albert graphs. We also developed a semi-analytical ´
model using a microscopic Markov-chain approach, theoretically derived key quantities, and established
strong alignment between these semi-analytical results and stochastic Monte Carlo simulations. Our
work further examined how various parameters like transmission rate, risk perception and peer influence
drive long-term epidemic outcomes.
Therefore, the primary objective of this thesis will be to study the evolution of epidemics through
two main lenses, firstly population diffusion among communities and secondly, evolution of opinions
on vaccination. Both these phenomenon actively affect epidemic outcomes, and developing a better
understanding and mathematical models to capture their behaviour helps towards the ultimate goal of
designing more robust and realistic intervention strategies for epidemic control and prevention in a
deeply interconnected world.

Abstract

Electrical current is one of the most accessible quantities to measure in an electrical system, and the
time-varying current drawn by an appliance often encodes a signature which can provide insights into its
operation. This thesis explores the use of current signatures, captured using low-cost Internet of Things
(IoT) hardware, in serving as a practical proxy for monitoring utilities in settings where conventional
instrumentation is too expensive or operationally inconvenient. The work focuses on two problems
that are common in developing regions: (i) Water flow monitoring and (ii) Downtime detection and
behaviour analysis of Uninterruptible Power Supply (UPS) units.
For water networks, the thesis addresses the lack of continuous, automatic water flow monitoring
due to the widespread reliance on analog meters and the high cost of digital flow meters for larger
pipelines. It proposes an IoT device that measures current consumption of water motors and, combined
with pipeline pressure, estimates water flow indirectly. The models are trained using ground-truth flow
measurements during an initial calibration phase, after which flow can be estimated indirectly. The system design, data acquisition pipeline, and machine learning models are described, and their performance
is evaluated using long-term deployments with large datasets, along with testing against erasures in the
dataset.
For UPS monitoring, the thesis targets the limitations of vendor-specific, network-dependent solutions that fail to provide comprehensive data during outages. It proposes an Original Equipment
Manufacturer (OEM)-agnostic IoT device for non-intrusive sensing of UPS input and output currents
that continues to operate even when mains power and local network connectivity are unavailable. Algorithms are introduced that use current profiles to detect outages and infer key UPS operating states.
These methods are validated through multi-site field deployments and an accompanying dashboard for
downtime analysis. Adoption of this UPS monitoring system is under way at IIIT Hyderabad.
Across both case studies, the results show that current-based IoT sensing can provide reliable and
fine-grained visibility into water flow and UPS behaviour using inexpensive hardware and without intrusive modifications to existing infrastructure. This positions current-centric IoT sensing as a viable,
scalable alternative to conventional utility monitoring solutions in resource-constrained environments.

Abstract

Household assistive robots are increasingly expected to perform complex, multi-step daily activities in dynamic human environments. Traditional planning systems often focus on executing a single task at a time, which limits efficiency, adaptability, and opportunities for collaboration. Recent advances using Large Language Models (LLMs) show promise in enabling robots
to reason at higher levels of abstraction, but many existing approaches require substantial training data or remain disconnected from fine-grained execution. This thesis explores how LLMs
can be integrated with symbolic planning and structured knowledge to enable agents that can
anticipate future tasks, collaborate with humans, and refine their understanding of new tasks
with minimal supervision.
The first study presents a framework that leverages a small number of LLM prompts to anticipate upcoming high-level tasks in a household environment. These anticipated tasks are then
used as goals in a classical planner, enabling the robot to compute efficient, low-level action
sequences that jointly achieve multiple tasks. Compared with a non-anticipatory baseline, this
approach demonstrates significant improvements in execution time and overall task efficiency.
The second study extends this idea toward human-robot collaboration. An enhanced planning
system jointly reasons over anticipated tasks and produces coordinated action sequences for
both the human and the agent. The system is capable of adapting on the fly to unexpected
changes in human behavior, uncertainty, or action outcomes, resulting in improved fluency and
effectiveness of collaboration in simulated household settings.
The third study addresses the challenge of adapting an agent to new tasks with limited labeled
data or prior examples. The framework integrates LLM predictions with domain knowledge
stored in a Knowledge Graph (KG), enabling the agent to evaluate and revise proposed action
sequences based on environment and capability constraints. Importantly, the agent can expand
and refine its knowledge base on the fly, starting from very sparse initial knowledge. It does
this by asking targeted questions to humans when required, thus closing the loop between
LLM reasoning, structured knowledge, and human input. Experiments in simulated cooking
and cleaning environments demonstrate substantial performance improvements compared with
relying solely on LLM predictions.Together, these studies showcase a principled approach for developing assistive household
agents that are capable of anticipation, collaboration, adaptation, and continuous refinement
of their knowledge for improved task execution.

Abstract

This research explores how structural and network topologies influence phase transitions in systems
characterized by cyclic dominance. Such dynamics are essential for understanding biodiversity and
stability in biological and ecological contexts. We focus on two primary inquiries that are discussed
below.
First, we examine a novel Rock-Paper-Scissors (RPS) model structured as a doublet chain. By
applying the antisymmetric Lotka-Volterra equation (ALVE), we analyze how mass is redistributed
across the network as a function of the interaction rates. Our results show that the system exhibits
a distinct shift in steady-state behavior: mass concentrates at specific corners or edges of the chain
and decays exponentially into the bulk. This localization remains robust against small variations in
the rate parameters. We confirm the nature of these distinct states as phase-transitions using various
analyses, providing a link between classical population dynamics and phenomena seen in condensed
matter physics.
Second, we investigate the behavior of ordered phases in a hybrid Potts-RPS model on complex
networks. We contrast the behavior of scale-free Barabasi-Albert (BA) networks against homogeneous ´
Erdos-R ˝ enyi (ER) graphs to determine how structural heterogeneity impacts consensus. Through Monte ´
Carlo (MC) simulations, we find that while cyclic dominance drives disorder, the presence of greater
connectivity in networks acts to stabilize the ordered phase. Furthermore, we perform a comparative
benchmark of homogeneous mean-field (HMF) and degree-based mean-field (DMF) theories, characterizing the parameter regimes where these deterministic approximations accurately predict global ordering
and identifying the limits of their applicability in capturing the stochastic fluctuations.
The significance of this thesis is in addressing present research gaps in the field of topology and evolutionary game dynamics for modelling biological phenomena. While ALVE in 1D chains have been
shown to demonstrate boundary mass localization, systematic extensions to higher-dimensional lattices
remain limited. Furthermore, while cyclic dominance in Potts models and RPS games on heterogeneous
networks have been studied in isolation, this project proposes an integration of the Potts formalism with
RPS dynamics on heterogeneous networks to explore their combined effects. By integrating these frameworks, this thesis provides a comprehensive account of how spatial and connectivity-based structures
govern the robustness and self-organization of biological systems.