A bit about me ...

My broad interests lie in understanding the limits of learning by trying to answer questions such as when is learning possible and when is it not under certain model assumptions? Such questions are particularly interesting when posed in constrained settings, for instance, constraints on computational complexity or adversarial robustness. Understanding such fundamental limits foster the design of provably efficient and reliable Machine learning methods.

More recently, I have been fascinated with statistical and causal properties of interpolating estimators in overparameterized model classes. You can find our recent work on this topic here. For a general overview of my work, please see my publications.

Brief Bio:

I am a Ph.D. candidate in the International Max Planck Research School for Intelligent Systems. I am jointly supervised by Prof.Debarghya Ghoshdastidar at the Theoretical Foundations of Artificial Intelligence group, Technical University of Munich and Prof.Dr.Ulrike von Luxburg at the Theory of Machine learning group, University of Tuebingen.

Research interests

  • Theory of over-parameterized learning,
  • Causality
  • Kernel machines,
  • Concentration inequalities,
  • Statistical vs computational limits of learning.


  • PhD in Statistical learning theory, since December 2018.

    University of Tuebingen, International Max Planck Research School for Intelligent Systems.

  • Masters in Intelligent Adaptive Systems

    University of Hamburg.

  • Masters in Mathematics, Bachelors in Mechanical Engineering

    Birla Institute of Technology and Sciences, Pilani.

Recent Publications

Interpolation and Regularization for Causal Learning

We study the problem of learning causal models from observational data through the lens of interpolation and its …

Causal Forecasting - Generalization Bounds for Autoregressive Models

Despite the increasing relevance of forecasting methods, the causal implications of these algorithms remain largely unexplored. This is …

Learning Theory Can (Sometimes) Explain Generalisation in Graph Neural Networks

In recent years, several results in the supervised learning setting suggested that classical statistical learning-theoretic measures, …

Graphon based Clustering and Testing of Networks - Algorithms and Theory

Network-valued data are encountered in a wide range of applications and pose challenges in learning due to their complex structure and …

Recovery Guarantees for Kernel-based Clustering under Non-parametric Mixture Models

Despite the ubiquity of kernel-based clustering, surprisingly few statistical guarantees exist beyond settings that consider strong …



Research intern

Amazon DE

Sep 2020 – Present Tuebingen, Germany
During this internship, I was fortunate to work with Dominik Janzing in the Causality team. We theoretically investigated the question of when good statistical forecasting models also make good causal models. You can find the paper we wrote on this topic here.

PhD Candidate

International Max Planck Research School for Intelligent Systems

Dec 2018 – Present Tuebingen, Germany
The core of my work constitutes making progress towards understanding the fundamental limits of learning problems - answering questions such as when is learning possible and when is it not - and characterizing the interplay between statistical and computational properties of various learning algorithms. In addition, I explore tangential areas of interest such as understanding the generalization properties of algorithms. Please see my projects for a detailed overview of all my current projects.

Research intern

Max Planck Institute for Intelligent Systems

Jul 2018 – Nov 2018 Tuebingen, Germany
During the period of this internship, I investigated the theoretical properties of embeddings by answering the question, “What is the best possible sigma distortion and dimension that can be achieved by embedding arbitrary metric spaces (or doubling spaces) into Euclidian space (or more generally, $l_{p}$ spaces)”. In addition, I conducted an empirical analysis to answer the question, “Does sigma-distortion, unlike some of the other distortion measures, carry any predictive power as to whether an embedding is likely to achieve good generalization error in various machine learning tasks”. This work was published in NeurIPS 2019 titled “Measures of distortion for Machine learning”.

Teaching Assistant

Theory of Machine learning, University of Tuebingen

Apr 2017 – Jun 2018 Tuebingen, Germany
Provided support in conducting the lecture - Theory of Machine learning by Ulrike von Luxburg by creating and grading weekly assignments, conducting tutorial sessions and providing consultation hours.

Student assistant

Cognitive Systems group, University of Hamburg

Oct 2015 – Mar 2017 Hamburg, Germany
This project was a collaborative initiative between the Center for Manuscripts Analysis and the Cognitive Systems group. It involved creating a web-based, interactive image processing tool from scratch to aid in the analysis of Manuscripts for tasks such as writer identification, style transfer, and user-assisted document segmentation. In addition to creating an interactive, touch-enabled, front end tool for handling manuscripts, several computer vision algorithms were implemented on a C++ based server to achieve real-time performance. This allowed users to easily create custom pipelines and quickly run experiments for the aforementioned tasks.

Research assistant

Knowledge Technology group, University of Hamburg

May 2015 – Oct 2015 Hamburg, Germany
Created simulated interactive environments for robot testing for the tasks of object tracking, sound localization and gesture recognition.

Student assistant

CINACS group, University of Hamburg

Nov 2014 – May 2015 Hamburg, Germany
Visualization is still one of the fundamental tools we have to comprehend data. This project was aimed at providing a “visualization” tool for visually impaired individuals by means of haptic line graphs with additional verbal feedback. My task was creating various such haptic maps that were used in human experiments to determine when and what kind of additional auditory feedback would be helpful to better comprehend second-order information of such graphs.


  • Maria von Linden strasse 6, Tuebingen, 72076.
  • 30-5/A11, Theory of Machine learning group, Tuebingen AI research building.
  • Email Me