Abstract

Puns, as a form of wordplay, play a significant role in humor, language comprehension, and cultural expression. They rely on phonetic and semantic ambiguities to create humor and are widely used in entertainment, advertising, and literature. While computational approaches to pun generation and detection have advanced in English, pun generation in code-mixed languages remains largely unexplored. Code-mixing, the blending of linguistic elements from multiple languages within a single utterance, presents unique challenges due to its syntactic and semantic complexities. As multilingual communication continues to evolve, developing computational models that can understand and generate humor in such settings is increasingly important. This thesis presents a comprehensive study on the generation of puns in Hindi-English code-mixed text. We begin by constructing a resource of pun-alternate word pairs, which serve as the foundation for pun generation. These pairs are collected by experimenting with various phonetic similarity matching strategies designed to identify humorously interchangeable words across Hindi and English. Using these pairs, we generate puns and analyze their linguistic plausibility and humor potential. We evaluate several pre-trained multilingual language models for their effectiveness in generating syntactically and semantically coherent code-mixed text. Based on these insights, we propose several novel structured prompt-based methods for generating puns in the Hindi-English code-mixed setting. To assess the quality of generated puns, we design a comprehensive human evaluation framework and collect detailed annotations across multiple dimensions of humor and fluency. Human evaluation suggests that our proposed methods significantly outperform baseline approaches in terms of humor quality and contextual relevance. The annotated outputs from our evaluations form the Hindi-English Code-mixed Pun (HECoP) dataset, comprising 2000 human-annotated sentences. We leverage this dataset to compare various multilingual models on the task of pun detection. This resource provides a valuable benchmark for future research in multilingual pun generation and humor detection. Beyond pair-based methods, we further introduce a structured pun generation pipeline capable of generating puns from a single input word without relying on predefined pun-alternate lists. This pipeline integrates phonetic similarity analysis, compatibility scoring, and sentence filtering to enhance the coherence and humorousness of generated content. To the best of our knowledge, this is one of the first comprehensive computational studies focused on pun generation in code-mixed or low resource setting. The methodologies, evaluation frameworks,and datasets introduced in this work lay a strong foundation for future advancements in computational humor and multilingual NLP.

Abstract

Large Language Models (LLMs) have demonstrated remarkable capabilities across diverse domains, yet they exhibit critical inconsistencies that fundamentally undermine their reliability in real-world applications. This thesis addresses two fundamental challenges in LLM reasoning: the evaluation and enhancement of consistency in moral and logical reasoning tasks. We first tackle moral consistency evaluation, where traditional accuracy-based methods fail due to the subjective nature of moral reasoning. We introduce SaGE (Semantic Graph Entropy), an information-theoretic framework that quantifies moral consistency by analyzing the semantic coherence of underlying “Rules of Thumb” (RoTs) inferred from LLM responses. To support this evaluation, we construct the Moral Consistency Corpus (MCC), containing 50,000 moral reasoning instances across diverse scenarios. Our comprehensive evaluation reveals widespread moral inconsistencies across state-of-the-art LLMs, with the maximum observed SaGE score being only 0.681, indicating substantial reliability concerns. We then investigate logical consistency, focusing on the pronounced difficulties LLMs encounter when reasoning with counterfactual premises that conflict with their parametric knowledge. Through the CounterLogic benchmark, a systematically designed dataset spanning 9 formal inference schemas, we demonstrate substantial performance degradation (27% on average) when models reason against their parametric knowledge compared to knowledge-consistent scenarios. To address these logical consistency challenges, we propose Self-Segregate, a metacognitive intervention inspired by human cognitive strategies for handling conflicting information. This two-phase prompting technique first assesses the factual alignment of premises before performing logical reasoning, enabling epistemic compartmentalization. Self-Segregate significantly reduces counterfactual reasoning performance gaps from 27% to 11% while improving overall logical accuracy by 7.5% across multiple models and tasks. Our findings establish consistency as a critical dimension of LLM performance that is orthogonal to accuracy, revealing that models can achieve high task performance while remaining fundamentally unreliable. This thesis contributes essential methodologies for developing more robust and trustworthy language models through novel evaluation frameworks, systematic benchmarks, and effective intervention strategies.

Abstract

Cinema, as a pervasive cultural medium, wields profound influence over societal perceptions, yet its role in perpetuating objectification remains underexplored in non-Western contexts. This thesis bridges this gap by investigating how cinematic techniques in Indian ”item songs”—a genre marked by provocative choreography and strategic camera framing—to induce objectifying gaze behavior in viewers. Integrating this with computer vision model, we present a dual-methodological approach: an empirical eye-tracking study and a novel multi-modal deep learning framework for understanding and detecting visual objectification in videos. In the experimental component, 91 participants viewed sexualized (SV) and non-sexualized (TV) music videos while their gaze metrics—fixation duration, visit counts, and scanpaths—were recorded. Results revealed that sexualized framing significantly redirected attention toward objectified body regions (torso, lower body), with gaze synchronization rates 6× higher in SV than TV (p < 0.001). Dynamic segmentation and ScanGraph analyses demonstrated that camera techniques such as close-ups and rapid editing overrode individual differences, homogenizing gaze patterns across viewers. These findings empirically validate theoretical frameworks like objectification theory (Fredrickson & Roberts, 1997) and the male gaze (Mulvey, 1975), highlighting how exogenous cinematic cues force sexual gaze objectification. Complementing this, our computational contribution introduces an interpretable multi-modal AI framework. By fusing video (LLaVA-NeXT-Video-7B-hf), audio (Whisper-large-v2), and text (allmpnet-base-v2) embeddings via contrastive learning, the model quantifies objectification intensity by dynamically weighting multi-modal cues, achieving state-of-the-art objectification detection (F1: 0.783, Acc: 0.826). A concept bottleneck mechanism further links predictions to human-interpretable cinematic elements (e.g., ”male gaze framing,” AUC: 0.803). This work advances interdisciplinary research by quantifying the cognitive impact of cultural media practices and providing scalable tools for bias detection caused by directors. Its implications extend to AI-driven content moderation, policy frameworks, ethical cinematography, and cross-cultural studies of media effects, establishing a foundation for mitigating objectification in increasingly visual digital ecosystems.

Abstract

In this thesis, we study the problem of Security and Decentralization in distributed systems when incentive driven players are present. The challenge of security is crucial in distributed systems to ensure agreement of the current state among different players distributed across the globe. Distributed systems is used for critical infrastructures such as communication, finance and banking, data warehousing and storage. Therefore, the presence of security vulnerabilities allows for an attacker to exploit such infrastructure and cause disruptions — which can indirectly lead to severe consequences. Similarly, decentralization is another critical requirement from democratized distributed systems such as Blockchains, to ensure that no single player or a small set of like-minded players hold disproportionate control over the system. Centralization of a democratized distributed system such as blockchains can cause challenges such as censorship, reversing transactions, double payment, unilateral changes in the protocol, etc. Therefore, the challenges of security and decentralization of distributed systems are of critical importance. Recent literature has tackled this problem against a set of completely disruptive player (called Byzantine player) and a set of honest (altruistic) players that follow the protocol honestly. However, a more realistic model of the system would involve a discussion on incentive-driven players, which will follow a profit-maximizing strategy. This is the behavior of real-world players involved in such distributed protocols. Decentralization in Proof-of-Work (PoW) based blockchain protocols (which is a type of distributed consensus protocol) is challenged by Rational players – who aim to reduce risk in rewards through the formation of mining pools — where a group of miners give their mining power to a single pool manager in exchange for more frequent payments. This causes centralization of such PoW blockchains. Among existing works that address the problem of security in the presence of rational players, we cover the gaps and show that these bounds can be improved. We also show that a higher level of security is achievable through a clever design of the protocol. On the decentralization front, existing work aims to reduce centralization in PoW blockchains through protocol design. However, we propose a change in the class of reward mechanisms used in PoW blockchains that eliminates any incentives of centralization through pool formation. Overall, this thesis studies and solves the problem of security and decentralization in distributed systems. We identify that accountability and penalty are crucial components in designing such a mechanism, and show that we can improve the adversarial tolerance bounds by designing a consensus/agreement protocol using these two properties, while also guaranteeing security. In decentralization of PoW blockchains, we categorize the Block Reward Mechanisms (BRMs) into two categories — Memoryless and Retentive BRMs. We show that most existing blockchains use Memoryless BRMs and it is impossible to achieve true decentralization using such BRMs. We then propose our own retentive BRM — decentBRM which achieves true decentralization. This proves through construction that true decentralization can be achieved using Retentive BRMs. Our work shows changes in reward/penalty schemes can achieve a higher level of security and decentralization in the presence of rational players. The goal of this thesis is to demonstrate the possibility of solving challenges in cryptographic protocols involving rational players through changes in reward/penalty schemes.

Abstract

The Internet of Things (IoT) has transformed how devices interact, enabling seamless data exchange. However, traditional terrestrial networks that rely on cellular towers, Wi-Fi, and LPWAN technologies can only reach areas with existing infrastructure. This limitation leaves vast rural, remote, and oceanic regions without reliable connectivity, restricting the full potential of IoT applications. To overcome this challenge, satellite-based IoT (Sat-IoT) has emerged as a promising solution, extending connectivity beyond the reach of ground-based networks. By leveraging Low Earth Orbit (LEO) satellite constellations, Sat-IoT enables direct communication between IoT devices and satellites, ensuring global coverage. This thesis explores the performance of Sat-IoT networks using stochastic geometry, a mathematical framework that models the random placement of satellites and IoT devices in large-scale environments. The study focuses on coverage probability and meta-distribution, providing deeper insights into network reliability under different channel conditions. This thesis considers a direct-access LEO satellite-IoT architecture where IoT devices communicate directly with multiple satellites. The satellites selectively decode and forward the information with dedicated orthogonal resources, avoiding interference. Signals from each satellite are coherently combined at the GS using maximal ratio combining (MRC). The satellites are considered to be distributed at a fixed altitude around Earth following a Binomial point process. Derived generalized closed-form expressions for coverage and Meta-distribution(MD). The coverage and MD as a function of SNR threshold T, reliability threshold τ and the altitude rmin are analyzed. The results demonstrate the impact of mega-LEO constellations on coverage and reliability, guiding 6G architecture design to improve connectivity and data offloading in smart cities and dense urban environments and also highlights that the proposed architecture can achieve high coverage and reliability with 8-10 satellites at altitudes between 800 km and 1400 km, as seen in upcoming mega-LEO constellations like Starlink and OneWeb

Abstract

Natural Language Processing (NLP) has witnessed transformative progress with the advent of architectures based on transformers and their attention mechanisms. This thesis investigates the versatility and adaptability of attention networks across three increasingly complex tasks: fine-grained sentiment analysis, multilingual pronominal coreference resolution, and multimodal user-defined keyword spotting. By progressively applying attention architectures from single-modality, fine-grained classification to multilingual and multimodal problems, we establish a unified framework for sequence understanding and cross-domain transfer. We begin by exploring fine-grained sentiment analysis as a test bed to evaluate the ability of attention networks to model subtle emotional distinctions across multiple sentiment classes. Unlike coarse sentiment classification tasks, fine-grained sentiment analysis requires capturing nuanced affective expressions and their contextual dependencies. Using the English subset of “A Multilingual Dataset for Sentiment Analysis and Emotion Detection (XED)”, we train an attention-based architecture. Our model demonstrates a strong capacity to distinguish between emotions such as joy, anger, fear, and surprise, showing the efficacy of attention mechanisms in handling fine-grained multi-class classification. This phase serves as a foundational study of how attention models perform when semantic granularity is essential. Building on this, we investigate pronominal coreference resolution in a multilingual setting, aiming to determine whether attention networks can generalize across typologically diverse languages. Coreference resolution involves identifying which entities pronouns and noun phrases refer to within a discourse. We design an attention-based model capable of capturing long-range dependencies between mentions and their potential antecedents. Evaluated in a zero-shot transfer setup using our own multilingual coreference dataset, Multilingual GAP (mGAP), the model shows robust performance in resolving coreferences across multiple low-resource and morphologically rich languages, without requiring language-specific tuning. This demonstrates the potential of attention architectures in bridging linguistic variation and addressing the challenges of multilingual understanding. Finally, we extend and adapt the attention framework to the multi-modal task of user-defined keyword spotting, where the goal is to detect arbitrary text queries in continuous speech. Traditional keyword spotting systems rely on fixed vocabularies and specialized architectures, limiting their applicability. Our model uses a similar attention mechanism to align spoken utterances with textual queries in a flexible, vocabulary-agnostic manner. We evaluate our approach on the Google Speech Commands dataset and the Qualcomm Keyword Spotting dataset, introducing architectural optimizations to reduce inference latency while maintaining high retrieval accuracy, making the approach practical for real-time applications in resource-constrained environments. Through these three tasks, we highlight how attention networks can be systematically adapted for varied linguistic and multimodal challenges. Our findings emphasize the trade-offs between accuracy, generalizability, and computational efficiency, while underscoring the potential of cross-attention mechanisms to serve as a common foundation across diverse NLP tasks. This work contributes to the broader understanding of attention-based modeling and provides practical insights for designing adaptable neural architectures for real-world language technologies.

Abstract

Large Language Models (LLMs) often inherit biases from the web data they are trained on, which contains stereotypes and prejudices. These biases emerge due to the uncurated nature of web-scale datasets, reflecting societal stereotypes, historical imbalances, and implicit prejudices. As a result, LLMs risk perpetuating harmful biases in their outputs, leading to fairness concerns and ethical implications in real-world applications. Current research has sought to mitigate these biases by developing debiasing techniques and evaluation methods. However, the effectiveness of these debiasing approaches hinges on the quality of bias-benchmark datasets, which are used to measure and validate improvements. These benchmarks typically assess bias by observing an LLM’s behavior on a set of biased statements. However, these statements often lack contextual considerations, treating bias as a static attribute rather than a phenomenon that varies based on situational context. To address this limitation, we introduce a contextual reliability framework, which evaluates model robustness to biased statements by considering the various contexts in which they may appear. We argue that assessing bias in isolation—without incorporating situational factors—can lead to unreliable conclusions, as the same statement may be perceived differently in different contexts. To quantify this, we develop the Context-Oriented Bias Indicator and Assessment Score (COBIAS), a novel metric that measures the contextual reliability of biased statements by evaluating the variance in model behavior when context is added. Existing bias benchmarks suffer from several limitations. Since bias is inherently subjective, current datasets struggle to capture its full complexity. Statements in these benchmarks often lack sufficient context, leading to ambiguous or misleading assessments of model bias. Furthermore, there exists no systematic way to quantitatively evaluate these datasets. Our work aims to bridge this gap by proposing a structured approach to assess the contextual reliability of bias benchmarks, ensuring that they provide meaningful and interpretable measures of bias. By evaluating whether LLMs demonstrate consistent behavior across various situational contexts, our framework helps identify benchmark datasets that may be unreliable for assessing model bias. To evaluate our proposed metric, we augmented 2,291 stereotyped statements from two widely used bias-benchmark datasets (CrowS-Pairs and StereoSet) by adding contextual information. This augmentation process allowed us to systematically analyze how model responses change when relevant context is introduced. Our findings show that COBIAS aligns strongly with human judgment on the contextual reliability of biased statements, achieving a Spearman correlation of ρ = 0.65 (p = 3.4 × 10−60). This demonstrates that our metric can serve as a robust tool for refining bias benchmarks and improving the evaluation of debiasing techniques. By incorporating context-awareness into bias assessment, our work provides a step toward more reliable and interpretable bias mitigation strategies in LLMs. Our data and code are publicly available: https://github.com/priyanshul-govil/COBIAS

Abstract

Text summarization is a core NLP task, but traditional methods often fail to meet diverse user needs. This has led to a growing focus on controllable summarization (CTS), which allows users to guide the summarization process according to specific requirements. Despite this interest, a comprehensive survey detailing the various controllable attributes, their associated challenges, and existing solutions has been missing. This thesis addresses this gap by formally defining the CTS task and categorizing controllable attributes based on their characteristics and objectives. We provide an in-depth review of current datasets and methods within each category, identifying key limitations, and outlining future research directions in the field. Building on this foundational understanding of CTS, we then explore the less-developed area of Multi- Attribute Controllable Summarization (MACS). This thesis specifically investigates the MACS task using large language models (LLMs) and various learning paradigms, with a particular emphasis on low-rank adapters. We conducted experiments with different fine-tuning strategies to assess how effectively models can retain patterns linked to multiple controllable attributes. Furthermore, we propose a novel hierarchical adapter fusion technique designed to integrate knowledge from two distinct controllable attributes. Our findings, the challenges encountered, and suggestions for future advancements for MACS are presented.

Abstract

Graph Neural Networks (GNNs) are increasingly being used for a variety of ML applications on graph data. Because graph data does not follow the independently and identically distributed (i.i.d.) assumption, adversarial manipulations or incorrect data can propagate to other data points through message passing, which deteriorates the model’s performance. To allow model developers to remove the adverse effects of manipulated entities from a trained GNN, we study the recently formulated problem of Corrective Unlearning. We find that current graph unlearning methods fail to unlearn the effect of manipulations even when the whole manipulated set is known. We introduce a new graph unlearning method, Cognac, which can unlearn the effect of the manipulation set even when only 5% of it is identified. It recovers most of the performance of a strong oracle with fully corrected training data, even beating retraining from scratch without the deletion set, and is 8x more efficient while also scaling to large datasets. We hope our work assists GNN developers in mitigating harmful effects caused by issues in real-world data, post-training. Beyond introducing a novel method, this work advances scientific methodology in GNN unlearning. We first use adversarial evaluation for graph unlearning methods beyond privacy applications, showing that metrics must genuinely reflect unlearning efficacy in corrective settings. Our extensive baselining includes methods from other domains like image unlearning for the first time in GNN unlearning, revealing that non-graph-specific approaches can surprisingly outperform graph-specific SOTA. Furthermore, our rigorous ablations challenge prevalent assumptions in GNN unlearning literature; for example, we show that common practices like node unlinking are not universally beneficial.

Abstract

This thesis investigates the design, fabrication, and characterization of zinc oxide (ZnO)-based nanoparticle sensors, focusing on applications in gas sensing and microorganism detection. The research addresses critical challenges in environmental monitoring and medical diagnostics by introducing innovative, cost-effective, and portable sensor solutions. A microheater design optimized for precise thermal control was developed using joule and induction heating techniques, enabling reliable operation in microfluidic and biochemical systems. ZnO nanostructures, particularly nanorods, were explored for their unique properties, including high surface-to-volume ratios, biocompatibility, and thermal stability. A ZnO nanorod-based biosensor integrated with an electrochemical impedance spectroscopy (EIS) circuit demonstrated high sensitivity in detecting yeast concentrations, showcasing potential applications in biomedical diagnostics. To further enhance biosensing performance, functionalization of ZnO nanorods with Concanavalin A (ConA) protein was implemented, resulting in improved selectivity and sensitivity. Additionally, the development of flexible ZnO-based CO2 gas sensors on PDMS substrates highlighted their effectiveness in real-time gas detection for industrial and environmental safety. The integration of microheaters with gas sensors significantly improved their sensitivity and response by optimizing temperature conditions. Microheaters were utilized to enhance the chemical reactivity of ZnO nanorods, ensuring faster and more accurate gas detection, especially for CO2. The precise thermal control offered by the microheaters allowed sensors to operate efficiently in varying environmental conditions, making them highly suitable for both wearable applications and harsh industrial environments. The results underscore the scalability and efficacy of ZnO-based sensors, balancing affordability with performance. The proposed systems address limitations in existing technologies, such as limited portability and low specificity, and emphasize their suitability for IoT-enabled real-time monitoring. Future research can expand on these findings by exploring additional gas and pathogen detection capabilities, integrating advanced machine learning for data analytics, and enhancing the durability and energy efficiency of flexible and wearable sensors. These advancements will contribute to smarter, more sustainable monitoring systems across diverse domains.