Hello! I’m a faculty researcher (chargée de recherche) at Inria in Lyon, specializing in optimizing numerical computations with computer arithmetic tools. My focus includes enabling deep learning on resource-constrained devices and enhancing digital signal processing for real-time deployment.
I have joined Inria in 2023, after having served 4 years as an Associate Professor at University of Nantes. Before my academic role in Nantes, I gained valuable experience as a research resident at Intel Corporation in San Diego, USA, and as a postdoctoral researcher at Inria team PASCALINE in ENS Lyon. I did my PhD from 2014 to 2017 at Sorbonne University (UPMC) and LIP6 laboratory under the guidance of Christoph Lauter, Thibault Hilaire and Jean-Claude Bajard. Before that, I received a degree in Applied Mathematics at the Odessa National I. I. Metchnikov University in Ukraine.
Towards efficient and accurate software and hardware. Computer arithmetic approach.
My research topics are reliable computing and performance optimization of numerical computations for software (SW) and embedded hardware (HW). My focus is on the finite-precision data representations and how the choices related to the precision and architecture of arithmetic operations influence the SW/HW accuracy and performance. The use of fixed- or floating-point representations results in computational errors that can have a serious impact on the result of the execution. Furthermore, choices about arithmetic (e.g. precision data and of arithmetic operators) inevitably influence the performance of the software and/or the target hardware cost. The programmer’s burden is then to find a trade-off between the compute precision, the result’s accuracy and the performance.
My research philosophy for more than 10 years has been the following: developers are not required to be FP or FxP arithmetic specialists and we need to build tools that will automatically make architectural decisions to reach the “sweet spot” between performance and accuracy, using knowledge specific to the application domain.
My research has been gravitating around the design and optimization of arithmetic operators (from multiplication by constants to function approximation), and their use in digital signal processing and machine learning applications. Recently, I have been more involved in efficient deployment of AI models on resource-constrained devices.
You must be logged in to post a comment.