Vikram V Garg

San Francisco Bay Area
 Email: vikram.v.garg (at) gmail.com
CV

About

I build intelligent, predictive systems that combine data, domain knowledge and state-of-the-art software engineering practice.

Some examples of systems I have designed, built and provided technical leadership for:

  1. A new control and data collection system for direct joint-based control of robots. The new system allows operators to complete hitherto inaccesible tasks. New tasks provide a training pathway to improve the foundational models which underlie autonomous robot operation.
  2. A comprehensive, automated adjoint (backward propagation) calculation and deployment system in an advanced physics simulation library. The automated adjoint system has allowed researchers and practitioners to develop effective algorithms for design and inference in multi-physics, multi-model systems.
  3. A Bayesian experimental design system to help systematize a geophysical research group's heuristic based design space exploration practice.

I maintain a very significant interest in economics, specifically the questions of:
  1. How do new products and services emerge and become mainstream in large markets ? In other words, how is new, useful work created in economies ?
  2. How can economically challenged geographies and polities accelerate market expansion ? In other words, how do national/regional economies develop ?

Expertise & Accomplishments

  • Robot Learning Infrastructure
    • Designed and built systems for direct joint control of robots, and enable more advanced foundational model training.
    • Provided reliable and accessible robot simulation platforms to accelerate robotics research.
  • Optimization & Inference Platforms
    • Incorporated automated adjoint calculation and analysis algorithms in the advanced physics simulation library libMesh.
    • Adjoint based multi-model inference algorithms to guide model adaptivity and experimental design.
  • Statistics & Data Assimilation
    • Optimal experimental design strategies for inferring high-dimensional, nonstationary ocean model parameters using R-INLA.
    • Density estimation algorithms targeted at inferring heavy tailed distributions without using specialized heuristics.
    • First fully flexible incremental Latin Hypercube sampler enabling efficient high dimensional design space exploration.

Service

  • Co-organizer
    Mini Symposium on "Adjoints in Computational Software" - USNCCM 2017.

  • Reviewer
    SIAM Journal on Scientific Computing, Numerische Mathematik, Computers & Mathematics with Applications, Computer Methods in Applied Mechanics & Engineering, Numerical Methods for Partial Differential Equations, International Journal for Numerical Methods in Engineering.

Education

2007 - 2012

Ph.D. in Computational and Applied Mathematics
The University of Texas at Austin
Thesis: Coupled Flow Systems, Adjoint Techniques and Uncertainty Quantification
Advisors: Graham Carey, Serge Prudhomme

2003 - 2007

B.S. in Aerospace Engineering
B.S. in Pure Mathematics
The University of Texas at Austin
GPA: 3.97/4.00