Alexander Kolesnikov
Staff Research Scientist at Google Deepmind.
Pushing the frontier in image and multimodal models at Google Brain (now DeepMind) since 2018 in the beautiful city of Zürich. Training SOTA vision models (ImageNet SOTA in 2019, ImageNet SOTA 2020, ImageNet SOTA 2021) and SOTA open weight models: SigLIP and PaliGemma.
Along the way, inventing new architectures: BiT, ViT, MLP-Mixer and FlexiViT. Also trying to understand, unify and simplify deep learning: UViM, Vision with rewards, patient and consistent distillation, effective image-text pretraining with locked image tower or sigmoid loss. And much more in Google Scholar.
I also enjoy writing flexible and performant research infrastructure. Open-sourcing the research infra I write is even more fun: big_vision.
Previously, I did my PhD at ISTA under the supervision of Christoph Lampert, where I was working on weakly-supervised learning and generative image models.
Contact me at a@akolesnikov.ch.
