JSC Researchers Receive Outstanding Paper Award at NeurIPS 2022

Mehdi Cherti and Jenia Jitsev from JSC received an Outstanding Paper Award at NeurIPS 2022, one of the leading international conferences for machine learning and artificial intelligence, for their publication entitled "LAION-5B: An open large-scale dataset for training next generation image-text models". Current language-vision models have risen above traditional approaches and overcome numerous limitations. This is mostly thanks to training methods based on large-scale datasets from online sources. Unfortunately, research is mainly conducted by industry labs and datasets have remained private — until now. This award-winning work provides open datasets from public internet resources that can be used to train state-of-the-art language-vision models and that are accessible to all research labs around the world. The work was performed by JSC and LAION e.V., a Hamburg-based non-profit organization strongly committed to open source and open science that is co-led and was co-founded by Jitsev and Cherti. Toolsets for dataset composition and pre-trained openCLIP models are also open-sourced as a result of this work.

Contact: Dr. Jenia Jitsev

from JSC News No. 294, 7 February 2023