Benoit Rostykus

hello@benoitrostykus.com

I am a Machine Learning Engineer with a decade of experience building software for start-ups & large companies.

I currently work for Databricks on serverless compute products. Previously, I worked on Netflix cloud infrastructure.
At the intersection of machine learning and systems, I have built data-driven schedulers and predictive systems that make software run more efficiently and predictably, and control trade-offs between performance and costs.

I have expertise in large-scale convex optimization, sparse models, combinatorial optimization and computation in low-latency environments.

Links to public things I've worked on:

• Machine Learning for Systems [2020, NeurIPS, video], slides
• Predictive CPU isolation of containers at Netflix [2019, blog post], patent
• Beta Survival Models [2021, PMLR, paper]
• Causality without headaches [2018, slides]
• Large-scale Causal Learning [2017, NeurIPS, "What if?" workshop]
• vectorflow [2017, Github], blog post [2017], paper [2018, MLSys]
• an L-BFGS solver [2014]
• when I was in grad school, I had a tiny company selling this software [2010]

I graduated from Ecole Centrale Paris, Ecole Normale Supérieure de Cachan, and Lycée Louis-Le-Grand.
I live in San Francisco with my 2 young boys and wife.

My hobbies include being mediocre at all of the following: music production, embedded Linux & DSP, computational acoustics, CAD, electronics and woodworking.
My latest obsession is logic synthesis for ML hardware.

Other things:

• Multitenancy (Oct. 2023)