Mehdi Cherti
Postdoc at Jülich Supercomputing Centre (JSC), Germany

I received my PhD in Machine Learning from Université Paris Saclay in 2018 followed by a Postdoc at Mines ParisTech. Currently, I am a postdoctoral researcher at Jülich Supercomputing Center (JSC), within the Scalable Learning and Multi-Purpose AI (SLAMPAI) Lab. I am also part of the LAION team.
My research focuses on deep generative models and how to use them to learn representations that can help to design genuinely novel and useful objects, with various applications to computational creativity and scientific domains.
I am also generally interested in learning models that can learn efficiently (transfer learning, few-shot learning, meta-learning) and generalize out of their training distribution (OOD generalization), enabling broader and more robust applicability.
News
Feb 27, 2023 | Our “Reproducible scaling laws for contrastive language-image learning” paper was accepted to CVPR 2023! In the repo, we open-source all the models and we will provide soon as well all details about pre-training and evaluation details for reproducibility. |
---|---|
Nov 21, 2022 | Our “LAION-5B: An open large-scale dataset for training next generation image-text models” received an Outstanding paper award On NeurIPS Datasets and Benchmarks track. |
Oct 13, 2022 | Our “LAION-5B: An open large-scale dataset for training next generation image-text models” (announcement page, arXiv, poster ) paper was accepted at NeurIPS 2022 Datasets and Benchmarks track. |
Jul 1, 2022 |
Effect of large-scale pre-training on full and few-shot transfer learning for natural and medical images
|
Jun 9, 2021 | New preprint Effect of Pre-Training Scale on Intra- and Inter-Domain Full and Few-Shot Transfer Learning for Natural and Medical X-Ray Chest Images, where we conduct large-scale pre-training on large source datasets of either natural (ImageNet-21k/1k) or medical chest X-Ray images and compare full and few-shot transfer using different target datasets from both natural and medical imaging domains. Code is available here. |
Selected publications
- Reproducible scaling laws for contrastive language-image learningarXiv preprint arXiv:2212.07143 2022
- LAION-5B: An open large-scale dataset for training next generation image-text modelsIn Thirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmarks Track 2022
- Effect of large-scale pre-training on full and few-shot transfer learning for natural and medical imagesarXiv preprint arXiv:2106.00116 2021
- Optimization of classification and regression analysis of four monoclonal antibodies from Raman spectra using collaborative machine learning approachTalanta 2018
- The RAMP framework: from reproducibility to transparency in the design and optimization of scientific workflows2018
- Spurious samples in deep generative models: bug or feature?arXiv preprint arXiv:1810.01876 2018
- Digits that are not: Generating new types through deep neural netsIn Proceedings of the Seventh International Conference on Computational Creativity 2016