Loading...
Hello!
I’m Nikita Durasov, a Research Scientist at NVIDIA (
), working on semantic video search and vision-language models (VLMs). I build vector search and knowledge-based systems that transform massive video data into searchable, structured knowledge for advanced multimodal reasoning and scalable search.
Previously, I worked at Apple, Amazon, and Samsung AI on deep learning and computer vision. I completed my PhD at EPFL (🇨🇭) with Pascal Fua at CVLAB, focusing on uncertainty and robustness.
Research
I’m interested in advancing vision-language models (VLMs) and video understanding, building scalable retrieval and reasoning pipelines that turn complex video data into searchable, interpretable knowledge.
My work explores robustness, generalization, and semantic search to make multimodal AI systems more powerful and reliable. For a non-exhaustive list of my research interests, refer to my Google Scholar.
I have also contributed as a reviewer for various academic conferences and journals, such as: CVPR NeurIPS ICLR ICML ICCV ECCV ICRA IROS ACCV RA-L IJCV AISTATS TPAMI TMLR, and other (full list).
Publications
ZigZag: Universal Sampling-free Uncertainty Estimation Through Two-Step Inference
Nikita Durasov, Nik Dorndorf, Hieu Le, Pascal Fua
TMLR 2024 ICMLW 2024
openreview | pdf | code | page
Preprints
PartAL: Efficient Partial Active Learning in Multi-Task Visual Settings
Nikita Durasov, Nik Dorndorf, Pascal Fua
ARXIV 2022
Invited Talks
- 01/04/2025: "Test-Time Training methods for Generalization" was presented at AI Tinkerers in Lausanne, Switzerland. [demo]
- 13/02/2025: "Uncertainty and Generalization in CV" was presented at LauzHack DL Bootcamp in Lausanne, Switzerland. [slides] [video]
- 15/08/2024: "Deep Uncertainty" was presented at Mila - Quebec AI Institute in Montréal, Canada.
- 10/04/2024: "Deep Uncertainty" was presented at UiT The Arctic University of Norway in Tromsø, Norway.
- 12/02/2024: "Deep Uncertainty" was presented at MBZUAI in Abu Dhabi, UAE.
- 08/01/2024: "Low-cost Uncertainty Estimation for Deep Learning" at CVLAB Group Meeting in Lausanne, Switzerland.
- 02/10/2023: "Deep Uncertainty" was presented at Nordic AI Conference in Copenhagen, Sweden.
- 17/08/2023: "Deep Uncertainty" was presented at Stony Brook University in New York, USA.
- 16/08/2023: "Deep Uncertainty" was presented at NYU Courant Institute in New York, USA.
- 09/08/2023: "Deep Uncertainty" was presented at Apple's Park in Cupertino, USA.
- 08/08/2023: "Deep Uncertainty" was presented at Meta in San Francisco, USA.
- 13/12/2022: "Masksembles for Uncertainty Estimation" at Uncertainty Reading Group at ETH in Zurich, Switzerland. [video]
- 30/09/2022: "Enhanced Robustness of Transformer Detectors for Handling Omissions in Ground Truth Data" at Apple Development Center in Zurich, Switzerland.
- 26/08/2022: "Single-shot Uncertainty Estimation" at CVLAB Group Meeting in Lausanne, Switzerland.
- 26/04/2022: "Partial Active Learning with Incomplete Annotations" at VILAB Group Meeting in Lausanne, Switzerland.
- 20/12/2021: "Robust Prediction of Stereo Disparity Under Uncertainty" at Amazon Prime Air in Graz, Austria.
- 22/06/2021: "Masksembles for Uncertainty Estimation" at CVPR 2021. [video] [paper]