Spandan Ghoshal

Preprints, ongoing projects,
and areas of inquiry.

Broadly: nonparametric inference and sequential methods for dependent data — procedures that stay robust under weak dependence and heavy tails.

Focus

Current directions

§ i

Change-Point Detection

Nonparametric, distance-based methods for identifying distributional shifts in multivariate and high-dimensional data streams, with control over detection delay and false alarms.

Pairwise distancesBootstrap calibration
§ ii

Distribution-Free Testing

Multivariate two-sample problems approached through combinatorial and graph-based statistics, notably runs tests built on the shortest Hamiltonian path.

Runs testsConsistency theory
§ iii

High-Dimensional Statistics

Asymptotic behaviour of inference procedures when the dimension grows with or far exceeds the sample size, including HDLSS regimes.

HDLSS asymptoticsHeavy tails
§ iv

Random Graphs & Motifs

Estimation of motif counts and structural functionals in large random graphs, and the statistical properties of such estimators under different sampling models.

Motif estimationNetwork models

Papers

Preprints

  1. On High-Dimensional Change-Point Detection Based on Pairwise Distances

    Spandan Ghoshal, Bilol Banerjee, and Anil K. Ghosh

    Submitted to Statistica Sinica

    arXiv:2511.10078

In progress

Ongoing work

  1. A Graph-based Distribution-free Method for Change-point Analysis of High-dimension Data

    Spandan Ghoshal, Trisha Dawn, D. Kundu, Angshuman Roy, and Anil K. Ghosh

    In preparation, with collaborators at ISI Kolkata, Texas A&M University, Iowa State University, and IIT Tirupati

  2. Hierarchical Clustering for Multiple Change-Point Detection in High-Dimensions

    Spandan Ghoshal, Trisha Dawn, and Angshuman Roy

    In preparation

  3. On the Large-Sample Consistency of the Distribution-Free Multivariate Two-Sample Runs Test Based on the Shortest Hamiltonian Path

    Spandan Ghoshal and Anil K. Ghosh

    In preparation

Interested in collaborating?

If our research interests overlap, or you simply want to discuss an idea, I would be glad to hear from you.

Write to me →