Abstract: The Jacobi prior offers an alternative Bayesian framework for predictive modelling, designed to achieve superior computational efficiency without compromising predictive performance. This scalable method is suitable for image classification and other computationally intensive tasks. Compared to widely used methods such as Lasso, Ridge, Elastic Net, uniLasso, the MCMC-based Horseshoe prior, and non-Bayesian machine learning methods including Support Vector Machines (SVM), Random Forests, and Extreme Gradient Boosting (XGBoost), the Jacobi prior achieves competitive or better accuracy with significantly reduced computational cost. The method is well suited to distributed computing environments, as it naturally accommodates partitioned data across multiple servers. We propose a parallelisable Monte Carlo algorithm to quantify the uncertainty in the estimated coefficients. We establish that the Jacobi estimator is asymptotically close to, and asymptotically equivalent to, the posterior mode under the Jacobi prior.  To demonstrate its practical utility, we conduct a comprehensive simulation study comprising seven experiments focused on statistical consistency, prediction accuracy,  scalability, sensitivity analysis and robustness study. We further present three real-data applications: credit risk modelling using U.S. Small Business Administration (SBA) loan default data, multi-class classification of stars, quasars, and galaxies using Sloan Digital Sky Survey data, and spinal degeneration classification using sagittal MRI scans from the RSNA 2024 Lumbar Spine Degenerative Classification Challenge. In the spine classification task, we extract last-layer features from a fine-tuned ResNet-50 model and evaluate multiple classifiers, including Jacobi-Multinomial logit regression, SVM, and Random Forest. The Jacobi prior achieves state-of-the-art results in recall and predictive stability, especially when paired with domain-specific features. In addition, it substantially outperforms competing methods in runtime by large margins, leading to significant reductions in computational cost. This highlights its potential for scalable, high-dimensional learning in medical image analysis.

All code and datasets used in this paper are available at: \url{https://github.com/sourish-cmi/Jacobi-Prior/}

This is a joint work with my PhD student, Shouvik Sardar.

Abstract: Public institutions in India generate vast volumes of administrative and financial data across systems that are often designed for reporting and administration rather than analysis. These systems are usually fragmented across departments, platforms and databases, with differences in definitions, formats, reporting structures and levels of data quality. Therefore, the fault lines in the underlying data propagate through systems and ultimately distort the evidence used for policy decisions. This fragmentation is not incidental but structural, persists across organisations, and only a minority of public organisations report having a comprehensive operational view of their own data. Further, these fault lines do not stay contained within the systems that produce them, they propagate outward and distort the evidence base on which policy decisions are made. Systematic reviews of evidence-informed policymaking consistently identify such institutional deficits. This is further compounded by siloed decision-making and misaligned incentives across agencies, which often prevent the same data from being reused or reconciled even where the technical means exist. Lastly, analytical failure in evidence-based decision-making is rarely just the data problem; it is also a state capacity and capability issue. Layered on top of data design failure is the capacity failure. The individual-level competencies and organisational resources needed to convert raw data into usable evidence remain unevenly distributed within public institutions. These three fault lines, i.e., system design, analytical capability, and state capacity, are not independent failures to be addressed piecemeal but rather a single, connected chain running from system design to policy evidence.

Abstract: This talk shall be in the domain of Formal Methods. It shall briefly highlight one of the techniques in this area and a domain it is applied to. The technique is based out of ideas from symbolic simulation. And the area where it will be applied is to Error Correction coding schemes. Hope you guys will get to witness a day in the life of a Formal Verification Engineer -- the non insta version!

Abstract: We study a one-parameter family of zero-sum games arising from electoral competition with abstention. Two parties choose positions on a policy line; voters vote for the nearest party within a tolerance radius and otherwise abstain, while parties compete on plurality.

For general voter densities, the existence of a pure-strategy Nash equilibrium need not be monotone in the tolerance radius: an equilibrium may exist, disappear, and later reappear. We identify a geometric property of the winning correspondence, called centripetal structure, that behaves differently: once it appears, it persists as tolerance increases and guarantees a pure-strategy Nash equilibrium.

Preprint: (joint work with Aman Ray, BA, 3rd year, MSE) https://arxiv.org/pdf/2608.03788