DataPink
@data_pink
Solutions and expertise on coupling #GeoSpatial data with #Machine and #DeepLearning.
ID: 846608829932605440
https://data.pink 28-03-2017 06:24:09
78 Tweet
138 Followers
181 Following
Blogpost on our new theory for word2vec-like representation learning methods for images, text, etc. Explains why representation do well on previously unseen classification tasks offconvex.org/2019/03/19/CUR… Relevant to meta learning, transfer learning? Paper arxiv.org/abs/1902.09229
Other fundamental tool for non-smooth optim: subdifferential of f at x = set of all slopes of affine minorants of f which are exact at x. Coincides with {∇f(x)} if f cvx & differentiable. Leads to 1st order optimality condition for *non-smooth* cvx fcts Fabian Pedregosa Pierre Ablin
A walkthrough of the PyTorch Internals by core developer Edward Z. Yang . It's a great resource if you want to contribute to PyTorch. blog.ezyang.com/2019/05/pytorc…
Great news! We've open-sourced HighRes-net for Multi-Frame Super-Resolution by Recursive Fusion. github.com/ElementAI/High… Our #PyTorch implementation that topped AdvancedConceptsTeam's competition. We're hoping this will help researchers working with satellite imagery #AIforGood @element_ai
#ComputerVision ecosystem for GeoSpatial Imagery, at scale: @RoboSatPink datapink.com/presentations/… #NLP: From Text2Map, a state of art. datapink.com/presentations/… Both talks FOSS4G 2024 | https://fosstodon.org/@foss4g, on how to use #DeepLearning to extract valuable patterns from #GeoSpatial (open)data.
C'est la rentrée ! Le #meetup #PAU #MachineLearning ouvre ses portes le mercredi 11 Septembre à 18h30 Technopole Hélioparc . Au menu #ComputerVision, #DeepTech et #NLP avec Olivier Courtin from DataPink Inscription ICI : meetup.com/fr-FR/Meetup-M…
DataPink will be PARIS SPACE WEEK 2020 Plan to see us on our booth, for neat-EO.pink demo #AI4EO
WTF? We brutally dismember BERT and replace all his organs? 👉🏻 Check our latest work: arxiv.org/abs/2002.02925 ⛵️ BERT-of-Theseus ⛵️ *NEW MODEL COMPRESSION METHOD* *ONE* loss + *ONE* hyperparameter + *NO* external data = GREAT PERFORMANCE with a Hugging Face -compatible weights