Orkut
Search results
Cloob.com was a Persian-language social networking website, mainly popular in Iran. After the locally (and internationally) popular social networking website Orkut was blocked by the Iranian government, a series of local sites and networks, including Cloob, emerged to fill the gap.
CLOOB consistently outperforms CLIP at zero-shot transfer learning across different architectures and datasets. A sketch of CLOOB and zero-shot transfer learning results for CLOOB and CLIP (YFCC pretraining):
کلوب (به انگلیسی: cloob) وبگاهی ایرانی بود که در یکم دی ماه ۱۳۸۳ به عنوان یک شبکه اجتماعی راه اندازی شد و ۱۵ مرداد ۱۴۰۰ به فعالیت خود پایان داد. این تارنما امکان ارتباط ایرانیان جهت تبادل ...
Oct 21, 2021 · CLOOB is a novel method that uses modern Hopfield networks and InfoLOOB objective to improve zero-shot transfer learning. It outperforms CLIP, a foundation model for vision and language tasks, on various datasets.
- Andreas Fürst, Elisabeth Rumetshofer, Viet Tran, Hubert Ramsauer, Fei Tang, Johannes Lehner, David P...
- 2021
We propose to use the InfoLOOB objective to mitigate this saturation effect. We introduce the novel "Contrastive Leave One Out Boost" (CLOOB), which uses modern Hopfield networks for covariance enrichment together with the InfoLOOB objective.
“Cloob.com was entirely blocked three times and the last time it took 28 days to unblock it,” said company director Mohammad Javad Shakouri Moghadam in a blog post.
We introduce “Contrastive Leave One Out Boost” (CLOOB) which combines the “InfoLOOB” objective with modern Hopfield networks. CLOOB overcomes CLIP’s problems of (i) the saturation problem of the InfoNCE objective and (ii) the “explaining away” problem when the covariance structure is insufficiently extracted.
