Leena C Vankadara

Applied Scientist II, AGI Foundations@Amazon Research

leena.chennuru.vankadara [AT] gmail.com

Exciting News!

I am thrilled to announce that I will be joining the Gatsby Computational Neuroscience Unit at UCL as a Lecturer (Assistant Professor) in Feb 2025. This new role marks an exciting chapter in my academic career. Stay tuned for updates on my upcoming courses, research projects, and other activities at the Gatsby Unit!

I'm hiring Ph.D. students and Postdocs!

Prospective Ph.D. students: please apply here: Gatsby Unit PhD programme.

Prospective Postdocs: please get in touch via email!

Short Bio

I am an Applied Scientist at the AGI Foundations Lab at Amazon, where I am currently conducting fundamental research on the theory and science of scaling large language models. I previously earned my Ph.D., graduating summa cum laude, from the International Max Planck Research School for Intelligent Systems with Debarghya Ghoshdastidar at the Theoretical Foundations of Artificial Intelligence group, Technical University of Munich, and Ulrike von Luxburg at the Theory of Machine Learning group, University of Tuebingen. During my Ph.D., I spent 6 months at the Causality Lab of Amazon Research.

My research aims to develop efficient, reliable, and trustworthy machine learning models by understanding the theoretical foundations of deep learning and causal learning. Deep learning theory and causality present some of the most fascinating challenges in modern machine learning. Classical theories of statistical learning and optimization often struggle to explain the empirical success of deep learning algorithms or even provide reliable guidance on their behaviour. As models continue to scale to unprecedented levels, revealing emergent abilities, we need fundamentally new theories to explain their behavior. To tackle some of these questions, my research aims to i) develop a unified theory of scaling and ii) understand the learning dynamics and generalization in deep learning. Meanwhile, Causality remains in its nascency, with formal mechanisms still emerging. My research will address open problems in causality, including establishing robust formalisms for causal reasoning and developing frameworks to evaluate causal models.

Recent News

Advisees/Interns

Publications

Most recent publications on Google Scholar.
indicates equal contribution.

On Feature Learning in Structured State Space Sequence Models

Leena C Vankadara*, Jin Xu*, Moritz Haas, Volkan Cevher

ICML NGSM Workshop 2024 (Spotlight; Runner up for the best paper award)

Effective Sharpness Aware Minimization Requires Layerwise Perturbation Scaling

Moritz Haas, Jin Xu, Volkan Cevher, Leena C Vankadara

ICML HiLD Workshop 2024

Self-compatibility: Evaluating causal discovery without ground truth

Philipp M Faller, Leena C Vankadara, Atalanti A Mastakouri, Francesco Locatello, Dominik Janzing

AISTATS 2024

Reinterpreting causal discovery as the task of predicting unobserved joint statistics

Dominik Janzing, Philipp M. Faller, Leena C Vankadara

A preprint: Arxiv'2023

Explaining Kernel Clustering via Decision Trees

Maximilian Fleissner, Leena C Vankadara, Debarghya Ghoshdastidar

ICLR 2024

A Consistent Estimator for Confounding Strength

Luca Rendsburg, Leena C Vankadara, Debarghya Ghosdastidar, Ulrike von Luxburg

A preprint: Arxiv'2022

Interpolation and Regularization for Causal Learning

Leena C Vankadara, Luca Rendsburg, Ulrike von Luxburg, Debarghya Ghoshdastidar

NeurIPS'22: Neural Information Processing Systems. 2022.

Causal Forecasting - Generalization Bounds for Autoregressive Models

Leena C Vankadara, Philipp Michael Faller, Mila Hardt, Lenon Minorics, Debarghya Ghoshdastidar, Dominik Janzing

UAI'22: Uncertainity in Artificial Intelligence. 2022.

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

Leena C Vankadara, Sebastian Bordt, Ulrike von Luxburg, Debarghya Ghoshdastidar

AISTATS'21: Artificial Intelligence and Statistics. 2021 (Oral presentation; 3% of all submissions)

On the optimality of kernels for high-dimensional clustering

Leena C Vankadara, Debarghya Ghoshdastidar

AISTATS'20: Artificial Intelligence and Statistics. 2020

On Feature Learning in Structured State Space Sequence Models

Leena C Vankadara*, Jin Xu*, Moritz Haas, Volkan Cevher

ICML NGSM Workshop 2024 (Spotlight; Runner up for the best paper award)

Effective Sharpness Aware Minimization Requires Layerwise Perturbation Scaling

Moritz Haas, Jin Xu, Volkan Cevher, Leena C Vankadara

ICML HiLD Workshop 2024

Self-compatibility: Evaluating causal discovery without ground truth

Philipp M Faller, Leena C Vankadara, Atalanti A Mastakouri, Francesco Locatello, Dominik Janzing

AISTATS 2024

Reinterpreting causal discovery as the task of predicting unobserved joint statistics

Dominik Janzing, Philipp M. Faller, Leena C Vankadara

A preprint: Arxiv'2023

Explaining Kernel Clustering via Decision Trees

Maximilian Fleissner, Leena C Vankadara, Debarghya Ghoshdastidar

ICLR 2024

A Consistent Estimator for Confounding Strength

Luca Rendsburg, Leena C Vankadara, Debarghya Ghosdastidar, Ulrike von Luxburg

A preprint: Arxiv'2022

Interpolation and Regularization for Causal Learning

Leena C Vankadara, Luca Rendsburg, Ulrike von Luxburg, Debarghya Ghoshdastidar

NeurIPS'22: Neural Information Processing Systems. 2022.

Causal Forecasting - Generalization Bounds for Autoregressive Models

Leena C Vankadara, Philipp Michael Faller, Mila Hardt, Lenon Minorics, Debarghya Ghoshdastidar, Dominik Janzing

UAI'22: Uncertainity in Artificial Intelligence. 2022.

Graphon based Clustering and Testing of Networks - Algorithms and Theory

Mahalakshmi Sabanayagam, Leena C Vankadara, Debarghya Ghoshdastidar

ICLR'22: International Conference on Learning Representations. 2022

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

Pascal Esser, Leena C Vankadara, Debarghya Ghoshdastidar

NeurIPS'21: Neural Information Processing Systems. 2021

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

Leena C Vankadara, Sebastian Bordt, Ulrike von Luxburg, Debarghya Ghoshdastidar

AISTATS'21: Artificial Intelligence and Statistics. 2021 (Oral presentation; 3% of all submissions)

Insights into Ordinal Embedding Algorithms - A Systematic Evaluation

Leena C Vankadara, Michael Lohaus, Siavash Haghiri, Faiz Ul Wahab, Ulrike von Luxburg

JMLR. 2023

On the optimality of kernels for high-dimensional clustering

Leena C Vankadara, Debarghya Ghoshdastidar

AISTATS'20: Artificial Intelligence and Statistics. 2020

Measures of distortion for machine learning

Leena C Vankadara, Ulrike von Luxburg

NeurIPS'18: Neural Information Processing Systems. 2018

Vitæ