INSTITUTIONAL REPOSITORY
Welcome to the Institutional Repository (IR) of the Indian Statistical Institute (ISI). You can find articles published by researchers of the Institute, It also preserves and enables access to many other digital contents including Dissertation theses, Convocation addresses, Question papers, official records and the collections of special mention. However, you can request us to get the restricted materials you need for your research and development.
Communities in DSpace
Select a community to browse its collections.
Recent Submissions
On the Faithfulness of Language Model-based Data-to-Text Generation
(Indian Statistical Institute, Kolkata, 2026-11-03) Mahapatra, Joy
Data-to-Text Generation (D2T)—the task of transforming structured or semi-structured data such as tables, graphs, and meaning representations into readable and faithful natural language—has advanced substantially with the advent of language modeling, particularly with Large Language Models (LLMs). Despite remarkable improvements in language model-based D2T systems in terms of readability, maintaining faithfulness---defined in terms of informativeness and factual consistency---remains a persistent challenge. This thesis presents a systematic exploration of faithfulness in language model-based D2T, focusing on both its evaluation and enhancement, with particular attention to influential factors such as source–reference divergence.
The thesis first introduces a structure-aware encoder-decoder framework, based on pre-LLM architectures such as Long Short-Term Memory (LSTM), which encodes input data according to its record-field hierarchy and enhances informativeness through selective gating and hierarchical encoding. It then transitions to LLM-based D2T, conducting a large-scale evaluation across multiple D2T benchmarks and LLM families. The study identifies source-reference divergence---the semantic mismatch between input data and reference text---as a primary factor influencing readability, informativeness, and overall faithfulness. To address this issue, a divergence-weighted training approach is proposed, employing an adaptive loss function that significantly improves model performance. Focusing specifically on factual consistency, the thesis presents a detailed evaluation of widely used LLMs using both automatic and human assessments, with improvements quantified through the Average Rate of Change. The findings demonstrate that source-reference divergence is a major contributor to factual inconsistency. To mitigate this effect, a divergence-aware selective generation framework is introduced, enabling models to abstain from producing factually inconsistent outputs under conditions of high divergence, thereby improving factual consistency without substantial coverage loss. Finally, empirical and mathematical analyses of scaling behavior reveal that factual inconsistency in D2T follows an exponential---rather than a power-law---scaling with respect to model size, compute, and data volume, an observation validated through a statistical framework, \textit{VaCScal}.
Collectively, the findings of this research provide both conceptual and empirical insights into faithfulness in language model-based D2T, laying a foundation for the development of future models that are not only readable but also factually faithful.
Fourier Sketching: A Property Testing Perspective
(Indian Statistical Institute, Kolkata, 2026-10-29) Roy, Manmatha
Fourier analysis of Boolean functions is a fundamental tool with far-reaching applications in complexity theory, learning theory, and cryptography. In many natural settings, functions of interest exhibit Fourier sparsity: they can be well approximated using only a small number of significant Fourier coefficients. In this thesis, we study Fourier-sparse Boolean functions from the perspective of property testing, where the goal is to infer global structural properties of an unknown function using only a small number of queries to its truth table, and where the efficiency of an algorithm is measured by its query complexity. Originating in the context of PCP constructions and later evolving into an independent area, property testing provides a natural lens for understanding the intrinsic complexity of combinatorial objects. Our first contribution provides a refined analysis of a sketching-by-hashing framework that operates directly in the Fourier domain. While such an approach was previously explored by Gopalan et al. (2011), we significantly strengthen its theoretical guarantees by deriving sharper structural bounds for the hashing process. As a consequence, we obtain a query-efficient local list-correction algorithm for the Hadamard code. This tool yields several applications, including query-efficient testers for distribution-free Fourier sparsity, testing membership in subclasses of Fourier-sparse boolean functions, linear isomorphism testing with respect to a known Boolean function, and sumset size estimation for set functions. All of these problems are well studied in learning theory, cryptography, and additive combinatorics. Our second contribution introduces a sketching-by-hashing framework based on subspace restrictions, operating in the input domain of Boolean functions. By analyzing how functions behave under random affine subspace restrictions, we establish several structural results that lead to improved algorithms for testing Fourier sparsity. The same framework enables estimation of the heaviest Fourier coefficient up to small additive error, with query complexity depending only on the desired accuracy. These problems are relevant across several areas of computer science. A Fourier sparsity tester, for instance, can serve as a preprocessing step for learning algorithms over Fourier-sparse function classes, validating whether an unknown function belongs to the target hypothesis class; the ability to estimate the heaviest Fourier coefficient, on the other hand, can be leveraged to design distinguishers for symmetric-key ciphers. Taken together, our work strengthens the connection between sketching-by-hashing techniques and the property testing of Fourier-sparse Boolean functions, and advances the understanding of query complexity for learning and testing in the Fourier domain. We expect the algorithmic techniques developed here to find further applications across many areas of computer science.
M. Tech(CS) 1 st Year Mid Semester Exam 2025 - Quantum Computation
(Indian Statistical Institute, 2025-02-26) Indian Statistical Institute
B Stat 1 st Year Second Semester Exam 2024 - Statistical Methods II
(Indian Statistical Institute, 2025-02-27) Indian Statistical Institute
M. Tech(CS) 1 st Year Second Semester Exam 2024 - Statistical Methods
(Indian Statistical Institute, 2025-02-26) Indian Statistical Institute