RT @bumblebike@twitter.com

@AlanZucconi@twitter.com It’s in a binder that I need to find, will do ASAP for you. Another gem in there was

🐦🔗: twitter.com/bumblebike/status/

RT @bumblebike@twitter.com

@SwiftOnSecurity@twitter.com This is from a 1979 presentation. We are v slow learners, it seems.

🐦🔗: twitter.com/bumblebike/status/

RT @MLSimpsons@twitter.com

S04E02: @yudapearl@twitter.com and the autonomous school bus.

🐦🔗: twitter.com/MLSimpsons/status/

RT @florian_sikora@twitter.com

Graphe des relations entre informaticiens en France selon les invitations dans les jurys de thèse (entre directeur et juré) et les "communautés" associées

Pour zoomer et naviguer dedans :
easyzoom.com/imageaccess/0d23e twitter.com/coulmont/status/14

🐦🔗: twitter.com/florian_sikora/sta

Back to remote teaching tonight, and probably for the next three weeks. 😞 I still don't know how to deal properly with practical exercises in remote. Let's hope it's only until the end of the semester.

Funnily enough, when you do this, you also don't need the SVM (à la InterFaceGAN) to estimate latent directions. You can just use average vector arithmetics, i.e. compute difference of class centroids.

Show thread

The first preprint of our PhD student Perla Doubinsky, with M. Crucianu et H. Le Borgne.

We show that you don't need weird tricks to disentangle latent directions in GANs. If you know the attributes (or you can estimate them), balancing the contigency matrix is enough.

Show thread

Looking for semantic directions in your GAN latent space? Do you fear entanglement because your dataset is biased and your attributes correlated?

Well, don't: you just have to resample. 😉

⚖️Multi-Attribute Balanced Sampling for Disentangled GAN Controls
arxiv.org/abs/2111.00909

RT @MLSimpsons@twitter.com

Deep learner fitting a dataset, however the latent space is too small (2021).

🐦🔗: twitter.com/MLSimpsons/status/

RT @alexandra_gros@twitter.com

Message de rappel : La deadline pour transmettre son dossier pour la qualification c'est aujourd'hui à 16h ! twitter.com/alexandra_gros/sta

🐦🔗: twitter.com/alexandra_gros/sta

🧑‍💻💽 github.com/elias-ramzi/ROADMAP

Code has been released for ROADMAP, the smoothly differentiable and decomposable AP approximation for all things deep and ranking.

Give it a try and send Elias your feedback! 😉

Show thread

RT @nshaud@twitter.com

Do you like image retrieval? Do you wish that there would be better differentiable approximations of the average precision to train your models?

Well, say hello to ROADMAP 👋: arxiv.org/abs/2110.01445

Accepted to NeurIPS 2021. 🎀
First paper of our grad student Elias Ramzi !

🐦🔗: twitter.com/nshaud/status/1445

RT @MLSimpsons@twitter.com

S011E14: Lisa, in this house we obey the laws of c̶o̶n̶v̶o̶l̶u̶t̶i̶o̶n̶s̶ t̶r̶a̶n̶s̶f̶o̶r̶m̶e̶r̶s̶ patches!

🐦🔗: twitter.com/MLSimpsons/status/

Show older
Mastodon

Server run by the main developers of the project 🐘 It is not focused on any particular niche interest - everyone is welcome as long as you follow our code of conduct!