Profile Image
Vamshi Madala on Google Scholar Vamshi Madala on GitHub Vamshi Madala on LinkedIn Vamshi Madala on X

vamchowdary72[at]gmail

Education

  1. UCSB

    PhD

    • Understanding and designing DNN architectures from approximation theoretic perspective, with focus on efficincy and provable generalization.
    • Advisor: Prof. Shivkumar Chandrasekaran

  2. UCSB

    Masters

    • Thesis: A study of generalization in deep neural networks
    • GPA: 4.0/4.0

  3. IIT Roorkee

    Bachelors in ECE

    • Thesis: Low-cost display devices using nanoparticles
    • Advisors: Prof. Brijesh Kumar and Prof. Sanjeev Manhas
    • GPA: 8.09/10.0

Experience

  1. Amazon AGI

    Applied Scientist Intern

    • Trained a novel Mixture of Experts (MoE) architecture to reduce inter-node communication costs.

  2. Stealth startup

    ML Researcher

    • Part of the founding team, I led groundwork and architectural setup for ML based solutions, developing Vision, NLP, and speech-based models for tasks in the supply chain industry.

  3. Apple

    Software Intern

    • Developed physics based algorithms to improve the Fall Detection feature. Created data processing pipelines to efficiently handle hundreds of hours of time series data.

  4. UCSF

    Grad Researcher

    • Developed Medviz - an AWS web portal and visualization tool for deploying machine learning models for processing of large PET/MRI datasets.

  5. Briteseed

    ML Intern

    • Trained CNNs on hyperspectral image data from surgical tools to detect tissues.

  6. Samsung Research

    Engineer

    • Music Information Retrieval (MIR).

  7. VIOS Medical

    Embedded Intern

    • Characterized different wireless modules for energy consumption and connectivity.
    • Developed software packaging and Linux distribution tools.

I am a fifth year PhD student at University of California Santa Barbara (UCSB), advised by Shiv Chandrasekaran.

My PhD thesis focusses on approximation theoretic approaches to design neural network architectures that are efficient and have robust generalization. My most recent work involves efficient neural network based solvers for PDEs.

Posts

Learning LU Factorization using gradient descent

Learning LU Factorization using gradient descent

December 27, 2025

Formulating LU factorization of linear operators as a neural network training problem by representing L and U as structured weight matrices.

[Code]

Publications

MNO: A Multi-modal Neural Operator for Parametric Nonlinear BVPs

MNO: A Multi-modal Neural Operator for Parametric Nonlinear BVPs

Vamshi C Madala, N Govindarajan, S Chandrasekaran

AAAI Workshop on AI to Accelerate Science and Engineering (AI2ASE), 2026

[Paper] [Code]
CNNs Avoid the Curse of Dimensionality by Learning on Patches

CNNs Avoid the Curse of Dimensionality by Learning on Patches

Vamshi C Madala, S Chandrasekaran, J Bunk

IEEE Open Journal of Signal Processing, vol. 4, pp. 233-241, 2023

[Paper] [Code]
Understanding and Visualizing Generalization in UNets

Understanding and Visualizing Generalization in UNets

A Rajagopal, Vamshi C Madala, T.A Hope, P Larson

Proceedings of the Fourth Conference on Medical Imaging with Deep Learning, 2021

[Paper] [Code]
Predicting Generalization in Deep Learning via Local Measures of Distortion

Predicting Generalization in Deep Learning via Local Measures of Distortion

A Rajagopal, Vamshi C Madala, S Chandrasekaran, P Larson

arXiv preprint arXiv:2012.06969, 2020

[Paper]
A Study of Generalization in Deep Neural Networks

A Study of Generalization in Deep Neural Networks

Vamshi C Madala

Master's Thesis, University of California, Santa Barbara, 2021

[Paper]