Dmitri Goloubentsev

CTO | Head of Automatic Adjoint Differentiation at Matlogica

Speaker

Dmitri spent more than 17 years as a quant at Canadian and British banks. During that time, he experienced firsthand the limitations of existing high-performance computing tools, particularly for calculating risks using adjoint algorithmic differentiation (AAD). He left his last banking role to build the tools he had wanted to use himself. That decision led to MatLogica.

MatLogica’s AADC is a just-in-time compiler that makes high-performance computing accessible to developers working with object-oriented C++ and Python models. It records a model’s execution and compiles it into machine code, delivering reported speedups of 6x to 1,000x over the original implementation, while calculating first-order sensitivities at an adjoint factor below one. Major Wall Street and global banks use AADC in derivatives workloads involving more than $4 trillion.

The latest release introduces AADC Tracer, which turns a model execution into a record that can be inspected and traced back to the source code, helping users understand and validate what the model actually calculated.

Dmitri is the inventor of the technology behind MatLogica and co-author of multiple publications on AAD and high-performance computing in quantitative modelling.