Feb 2022 – Aug 2022 · 6 mos · Moscow, Moscow City, Russia

• - conducted academic research in Tabular Deep Learning resulting in the corresponding paper "Revisiting Pretraining Objectives for Tabular Deep Learning"

Sep 2021 – Feb 2022 · 5 mos · Moscow, Moscow City, Russia

• Sped up Yandex advertisement banners recommendations model, based on Transformer architecture, by 20 percent on long user histories and integrated the upgraded model into production
• Sped up data conversion into vector representations in production by 10 percent
• Researched exhaustively advanced model compression via distillation resulting in its integration into auto-training pipeline

Sep 2020 – Aug 2021 · 11 mos · Moscow, Moscow City, Russia · On-site

• Developed more than 5 viral features that can now be found in Lensa and Prisma Apps including a neural network that can predict and copy the photo processing applied to any photo, advanced portrait style-transfer networks, Instagram screenshot detection, etc.

Jun 2020 – Sep 2020 · 3 mos · Moscow, Moscow City, Russia · On-site

• Implemented a fast and efficient super resolution model for style transfer that made it possible for the first time to increase the resolution of stylized photos in Prisma app

Oct 2019 – Jan 2020 · 3 mos · Moscow, Moscow City, Russia · On-site

Jun 2019 – Sep 2019 · 3 mos · Moscow, Moscow City, Russia