• Density estimation with atoms, and functional estimation for mixed discrete-continuous data [Under Review, TAS]

  Brief description: Classical density (or density-functional) estimation assumes an underlying distribution absolutely continuous w.r.t. Lebesgue measure, leading to poor performance for data distributions with both continuous and discrete components. We show that a simple modification of existing nonparametric methods removes this assumption while preserving theoretical guarantees and improving empirical results.

• Post-detection inference for sequential changepoint localization [Under Review, JRSSB]

  Brief description: We address a fundamental but largely unexplored challenge addresses a fundamental yet underexplored question in sequential changepoint analysis: how can we perform valid inference on the changepoint, once a (potentially blackbox or heuristic) sequential detection algorithm has declared that a change has occurred? Specifically, we construct confidence sets for the (unknown) changepoint using only the data available up to a data-dependent stopping time, at which an arbitrary sequential detection algorithm stops and signals a change. To the best of our knowledge, our work provides the first general methodology for post-detection changepoint localization that is both assumption-lean and agnostic to the change detection method.

• Robust likelihood ratio tests for composite nulls and alternatives [Under Review, IEEE Trans. on Information Theory]

  Brief description: We first modify Huber's robust SPRT for simple nulls and alternatives to instead employ a test supermartingale. The martingale tools allow us to generalize the usual Huber contamination model to show that our sequential test retains type-I error validity even under sequentially adaptive contaminations under the null. With this building block in place, we extend our approach to handle both composite nulls and composite alternative hypotheses, a task which has not been undertaken earlier, and utilizes te idea of reverse information projection. In all cases, we analyze the growth rate under the alternative, which asymptotically converges the optimal non-robust growth rate.

• Saha and Ramdas' discussion of "Poisson-focus" by Ward, Dilillo, Eckley and Fearnhead [JASA, 2025]

  Brief description: In this discussion article, we demonstrate that a specific variant of the e-detector offers a promising alternative to the Poisson-FOCuS algorithm for online gamma-ray burst detection, providing a notable reduction in computation time while maintaining comparable detection efficiency.

• Testing exchangeability by pairwise betting [AISTATS, 2024. ORAL]

  Brief description: We address the problem of testing exchangeability of a sequence of random variables under the recently popular framework of testing by betting. We design a new (yet simple) game in which we observe the data sequence in pairs. Even though betting on individual observations is futile, we show that betting on pairs of observations is not. We prove that our game leads to a nontrivial test martingale, which is interesting because it has been obtained by shrinking the filtration very slightly. We show that our test controls type-1 error despite continuous monitoring, and is consistent for both binary and continuous observations, under a broad class of alternatives.

• Group-Feature (Sensor) Selection With Controlled Redundancy Using Neural Networks [Neurocomputing, 2024]

  Brief description: In this work, we present a novel embedded feature selection method based on a Multi-layer Perceptron (MLP) network and generalize it for group-feature or sensor selection problems, which can control the level of redundancy among the selected features or groups and it is computationally more efficient than the existing ones in the literature. Additionally, we have generalized the group lasso penalty for feature selection to encompass a mechanism for selecting valuable groups of features while simultaneously maintaining control over redundancy. We establish the monotonicity and convergence of the proposed algorithm, with a smoothed version of the penalty terms, under suitable assumptions

• Robust Classification of High-Dimensional Data Using Data-Adaptive Energy Distance [ECML PKDD, 2023]

  Brief description: In this work, we developed robust, computationally efficient, and tuning-free classifiers tailored for high-dimensional low sample size (HDLSS) data, using the concepts of angular distance and generalized energy distance. It is shown that they yield perfect classification in the HDLSS asymptotic regime (where sample size is fixed but the dimension approaches infinity), under some fairly general conditions.

Talks