Builders & open source
Vaishaal Shankar
Machine learning researcher; shared last author of DataComp-LM at Apple and a co-lead of OpenCLIP
Vaishaal Shankar is a machine learning researcher whose cited work includes dataset research and open-source CLIP training. At Apple, Shankar was one of the shared last authors of DataComp-LM (2024), a testbed for controlled dataset experiments with a standardized corpus, pretraining recipes, and a broad evaluation suite, and contributed to the data, evaluation, and responsible AI work on Apple's 2024 foundation language models report; the 2025 MobileCLIP2 paper notes that Shankar's contribution was done while at Apple. Shankar is an original author of OpenCLIP, an MIT-licensed open-source implementation of CLIP, and is named in its README among the people who lead current development of the repository.12345
At a glance
- Current role
- None confirmed in the sources
- Work listed
- 3
- Sources cited
- 8
- Profile reviewed
Research
Much of Vaishaal Shankar's research concerns how datasets shape model behavior. In 2019 Shankar co-authored "Do ImageNet Classifiers Generalize to ImageNet?", which rebuilt test sets for CIFAR-10 and ImageNet to check for overfitting to reused benchmarks, and in 2020 co-authored a study of how ImageNet models hold up under distribution shifts that arise in real data.67
Shankar is an author of DataComp (2023), a testbed for multimodal dataset experiments built around a candidate pool of image-text pairs from Common Crawl, and one of the shared last authors of its language-model counterpart DataComp-LM (2024). Shankar also contributed to the data, evaluation, and responsible AI work on Apple's 2024 foundation language models report and co-authored MobileCLIP2 (2025).8123
Open-source work
Shankar is one of the original authors of OpenCLIP, an open-source implementation of CLIP whose citation entry dates to July 2021. The README names Shankar with Ross Wightman, Romain Beaumont, and Cade Gordon as the people leading current development, and the code is released under the MIT license.45
Work and contributions
OpenCLIP Open-source code
Named in the README among the people who lead current development of the repository and among the original authors; the repository is released under the MIT license.45
DataComp-LM: In search of the next generation of training sets for language models Published research
One of the shared last authors, listed with Apple, of the 2024 DataComp-LM paper, which provides a standardized corpus extracted from Common Crawl, pretraining recipes, and 53 downstream evaluations for dataset experiments.1
Do ImageNet Classifiers Generalize to ImageNet? Published research
Co-author, with Benjamin Recht, Rebecca Roelofs, and Ludwig Schmidt, of a 2019 paper that builds new test sets for CIFAR-10 and ImageNet by following the original dataset creation process and measures how classifiers generalize to them.6
Timeline
- Co-authored "Do ImageNet Classifiers Generalize to ImageNet?"6
- Co-authored a study measuring robustness to natural distribution shifts in image classification.7
- Released OpenCLIP with collaborators, according to the repository's citation entry.4
- Co-authored DataComp, a testbed for multimodal dataset experiments.8
- Co-authored DataComp-LM as one of its shared last authors.1
- Contributed to Apple's report on its 2024 foundation language models.2
- Co-authored MobileCLIP2, on multi-modal reinforced training.3
Documented roles
The sources do not confirm a current role; these are the roles they document.
Companies in their history
- Apple 27 people in the wiki
Related people
- Fartash FaghriConnected to Apple
- David GrangierConnected to Apple
- Edouard GraveConnected to Apple
- Ronan CollobertConnected to Apple
- Craig FederighiConnected to Apple
- Douwe KielaLanguage models · Multimodal models
- Albert JiangLanguage models · Datasets and data
- Guilherme PenedoDatasets and data · Language models
- Junyang LinLanguage models · Multimodal models
- Yuanzhi LiLanguage models · Datasets and data
Suggest a correction or removal
Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.
References
8 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026
- DataComp-LM: In search of the next generation of training sets for language models (arXiv 2406.11794)
- Apple Intelligence Foundation Language Models (arXiv 2407.21075)
- MobileCLIP2: Improving Multi-Modal Reinforced Training (arXiv 2508.20691)
- mlfoundations/open_clip README (The Team)
- mlfoundations/open_clip LICENSE (MIT)
- Do ImageNet Classifiers Generalize to ImageNet?
- Measuring Robustness to Natural Distribution Shifts in Image Classification
- DataComp: In search of the next generation of multimodal datasets