Yatin Dandi

Computer and Communication Sciences

École polytechnique fédérale de Lausanne

I'm a fourth year Ph.D student at École polytechnique fédérale de Lausanne working on the theory of deep learning and statistical physics of computation under the supervision of Prof. Lenka Zdeborová and Prof. Florent Krzakala . Broadly, I'm interested in rigorous analysis of the asymptotics of systems involving a large number of interacting entities, especially those involving dynamics or aspects related to statistical efficiency and computational hardness. This spans a wide range of problems from graph coloring to deep neural networks.

Previously, I obtained a dual-degree (Btech-MTech) in Computer Science at IIT Kanpur, India.

During my time at IIT Kanpur and EPFL, I've had the fortune of working under an amazing set of supervisors, mentors, and collaborators (in no particular order): Yue M. Lu, David Gamarnik at MIT, Bruno Loureiro, Martin Jaggi and multiple members of the MLO lab, Piyush Rai, Abhishek Kumar, Vinay Namboodiri , Arnout Devos (PhD student under Prof. Matthias Grossglauser), and Arthur Jacot (Phd student under Prof. Clement Hongler).

A complete and up-to-date list of publications and preprints can be found on my Google Scholar page.

Interests

Statistical Physics of Computation
Theory of Deep Learning
Optimization

Education

BTech-Mtech (Dual Degree) in Computer Science and Engineering, 2017

Indian Institute of Technology, Kanpur

Publications

Grouped by current research themes. See Google Scholar for citation data and the newest updates.

Deep Learning Theory

Deep Learning as Neural Low-Degree Filtering: A Spectral Theory of Hierarchical Feature Learning
Yatin Dandi, Matteo Vilucchio, Luca Arnaboldi, Hugo Tabanelli, Florent Krzakala

HiLD Workshop, ICML 2026
Scaling Laws from Sequential Feature Recovery: A Solvable Hierarchical Model
Arie Wortsman-Zurich, Hugo Tabanelli, Yatin Dandi, Florent Krzakala, Bruno Loureiro

HiLD Workshop, ICML 2026
Deep Learning of Compositional Targets with Hierarchical Spectral Methods
Hugo Tabanelli, Yatin Dandi, Luca Pesce, Florent Krzakala

International Conference on Machine Learning (ICML), 2026
Provable Learning of Random Hierarchy Models and Hierarchical Shallow-to-Deep Chaining
Yunwei Ren, Yatin Dandi, Florent Krzakala, Jason D. Lee

Conference on Learning Theory (COLT), 2026
The Computational Advantage of Depth: Learning High-Dimensional Hierarchical Functions with Gradient Descent
Yatin Dandi, Luca Pesce, Lenka Zdeborová, Florent Krzakala

Spotlight, NeurIPS 2025
Fundamental Limits of Learning in Sequence Multi-Index Models and Deep Attention Networks: High-Dimensional Asymptotics and Sharp Thresholds
Emanuele Troiani, Hugo Cui, Yatin Dandi, Florent Krzakala, Lenka Zdeborová

International Conference on Machine Learning (ICML), 2025
A Random Matrix Theory Perspective on the Spectrum of Learned Features and Asymptotic Generalization Capabilities
Yatin Dandi, Luca Pesce, Hugo Cui, Florent Krzakala, Yue M. Lu, Bruno Loureiro

Oral, AISTATS 2025
Fundamental Limits of Weak Learnability in High-Dimensional Multi-Index Models
Emanuele Troiani, Yatin Dandi, Leonardo Defilippis, Lenka Zdeborová, Bruno Loureiro, Florent Krzakala

AISTATS 2025
Asymptotics of Feature Learning in Two-Layer Networks after One Gradient-Step
Hugo Cui, Luca Pesce, Yatin Dandi, Florent Krzakala, Yue M. Lu, Lenka Zdeborová, Bruno Loureiro

Spotlight, ICML 2024
The Benefits of Reusing Batches for Gradient Descent in Two-Layer Networks: Breaking the Curse of Information and Leap Exponents
Yatin Dandi, Emanuele Troiani, Luca Arnaboldi, Luca Pesce, Lenka Zdeborová, Florent Krzakala

ICML 2024
Online Learning and Information Exponents: The Importance of Batch Size and Time/Complexity Tradeoffs
Luca Arnaboldi, Yatin Dandi, Florent Krzakala, Bruno Loureiro, Luca Pesce, Ludovic Stephan

ICML 2024
Repetita Iuvant: Data Repetition Allows SGD to Learn High-Dimensional Multi-Index Functions
Luca Arnaboldi, Yatin Dandi, Florent Krzakala, Luca Pesce, Ludovic Stephan

Preprint, arXiv, 2024
How Two-Layer Neural Networks Learn, One (Giant) Step at a Time
Yatin Dandi, Florent Krzakala, Bruno Loureiro, Luca Pesce, Ludovic Stephan

Journal of Machine Learning Research (JMLR); invited talk, NeurIPS 2023 M3L Workshop
Universality Laws for Gaussian Mixtures in Generalized Linear Models
Yatin Dandi*, Ludovic Stephan*, Florent Krzakala, Bruno Loureiro, Lenka Zdeborová

NeurIPS 2023
Understanding Layer-Wise Contributions in Deep Neural Networks through Spectral Analysis
Yatin Dandi, Arthur Jacot

Preprint, arXiv, 2021

Mathematics of Spin Glasses and Random Structures

Rigorous Asymptotics for First-Order Algorithms Through the Dynamical Cavity Method
Yatin Dandi, David Gamarnik, Francisco Pernice, Lenka Zdeborová

Conference on Learning Theory (COLT), 2026
Minimum Number of Monochromatic Subgraphs of a Random Graph
Yatin Dandi, David Gamarnik, Haodong Zhu

arXiv, 2026
Sequential Dynamics in Ising Spin Glasses
Yatin Dandi, David Gamarnik, Francisco Pernice, Lenka Zdeborová

Preprint, arXiv, 2025
Asymptotics of Non-Convex Generalized Linear Models in High-Dimensions: A Proof of the Replica Formula
Matteo Vilucchio, Yatin Dandi, Cedric Gerbelot, Florent Krzakala

Preprint, arXiv, 2025
Sampling with Flows, Diffusion and Autoregressive Neural Networks: A Spin-Glass Perspective
Davide Ghio, Yatin Dandi, Florent Krzakala, Lenka Zdeborová

Proceedings of the National Academy of Sciences, 2024; NeurIPS 2023 Workshop on Diffusion Models
Maximally-Stable Local Optima in Random Graphs and Spin Glasses: Phase Transitions and Universality
Yatin Dandi, David Gamarnik, Lenka Zdeborová

arXiv preprint

Machine Learning, Optimization, and Generative Models

Data-Heterogeneity-Aware Mixing for Decentralized Learning
Yatin Dandi, Anastasia Koloskova, Martin Jaggi, Sebastian U. Stich

NeurIPS 2022 Workshop: Optimization for Machine Learning
Implicit Gradient Alignment in Distributed and Federated Learning
Yatin Dandi*, Luis Barba*, Martin Jaggi

AAAI Conference on Artificial Intelligence; FL-ICML Workshop, 2021
Generalized Adversarially Learned Inference
Yatin Dandi, Homanga Bharadhwaj, Abhishek Kumar, Piyush Rai

AAAI Conference on Artificial Intelligence; NeurIPS 2020 Workshop on Self-Supervised Learning
Model-Agnostic Learning to Meta-Learn
Arnout Devos*, Yatin Dandi*

Proceedings of Machine Learning Research; NeurIPS 2020 Pre-registration Workshop
NeurInt: Learning Interpolation by Neural ODEs
Avinandan Bose*, Aniket Das*, Yatin Dandi, Piyush Rai

Spotlight, NeurIPS DLDE Workshop, 2021
Jointly Trained Image and Video Generation using Residual Vectors
Yatin Dandi, Aniket Das, Soumye Singhal, Vinay P. Namboodiri, Piyush Rai

Winter Conference on Applications of Computer Vision (WACV), 2020

Contact

yatin.dandi@epfl.ch