And another one for today: an interesting paper that designs a method to make inference through sampling efficiently on GPU, with tensors:

arxiv.org/pdf/1806.08593v3.pdf

Very interesting paper, which says that you can provably make all local optima global by just adding a neuron next to the output neurons of a model !

It's yet another one in a series of papers that give partial explanatory theories about why deep learning works so well.

If anyone has the time to check the proofs, please let me know.

arxiv.org/abs/1901.00279

Would it be possible to merge the decentralized collaborative annotations of csarven.ca/dokieli-rww with the communication features of ? That would be great, and would enable us to have long discussions about a paper in the Fediverse, while at the same time pointing to some specific comments in dokieli... ?

An 11-years old paper in the Economist economist.com/leaders/2008/03/ described the evolution of the web, from initial "walled garden" to openness thanks to open standards; it suggested social media could follow the same evolution, but it didn't, at least so far, because of the lack of such standards. But isn't finally ActivityPub now filling up this role, and couldn't this old paper be finally correct ?

Here is a book chapter that addresses the many issues with the current centralized web, and proposes an alternative based on "Solid":
ruben.verborgh.org/articles/re

There are several similar ideas as with the Fediverse I believe, but is Solid yet another decentralized protocol or can we try and think about getting the best of both worlds to make the Fediverse progress ?

@Gargron that's how it should be... You don't need figures, you know lots of people have read it.

Pixelfed federates now! 🎉

Merry Christmas fediverse. #pixelfed

Si vous manquez de choses intéressantes à lire au coin du feu, nous vous recommandons cet article de @Etienne010101. Il y est question d’ouverture des données à @MEL_Lille. medium.com/datactivist/les-six

Contre le pouvoir des plateformes, établir une portabilité sociale des données – – S.I.Lex –

[...]
L’idée de départ consiste à sortir de la vision trompeuse selon laquelle des plateformes comme Facebook seraient de simples « hébergeurs passifs » pour identifier un « pouvoir de contrainte » imposé à leurs utilisateurs. (...) ce pouvoir s’exprime aussi par leur faculté à capter nos liens sociaux et à les retourner contre nous.
[...]

frama.link/hQpMjD7d

@rra I tried my own implementation with flask, but the send message just returns a 202 HTTP code: no idea what's happening at the mastodon side ! I guess I'll really have to install my own mastodon server just to have access to the logs; no other way... 😕

@rra
Hi rra... So your implementation of the tuto with flask looks pretty good ! But it's not finished, isn't it ? Did you manage to send a post to @darius ?

@Gargron @mariusor @rra

OK, it's working, it was indeed the MIME type of webfinger that needed to be application/json. Although I can't directly write the actor file URI as I thought I could, but I have to search for name@domain.

Thanks for your help !

I'm actually going to hire an engineer to develop a "scientific-resource oriented" AP server. Any hint greatly appreciated. Thanks !

@Gargron @mariusor

No, you're right, actually the webfinger file does is not cast as json MIME: that may be the catch indeed ! 🙂

@Gargron @mariusor Yes it does have
"type": "application/activity+json",
but even when searching directly for uldocs.atilf.fr/actor.json
mastodon returns 0 user

Thanks for your reply !

@mariusor @Gargron
Hi Gargron, I'm very sorry, but I'm having a hard time trying to get your "basic AP tuto" working. Could you please tell me what's wrong with my actor file here:
uldocs.atilf.fr/actor.json

It'll probably only take 1 minute from your time, thanks anyway :)

@mariusor Thanks for the reply ! I tried to add inbox, outbox as requested by AP (but not by mastodon afaik), no luck with , although seems to see the actor (but cannot display its profile)

It'd be nice to have a detailed reqs documents somewhere for mastodon, because I'm not fan of reading all the source code ;-)

Thanks !

@cwebber This blog post should be straightforward, but it's not ! 😠
What is really expected for an actor file to be accepted by Mastodon ??
When I search for uldocs.atilf.fr/actor
it returns 0 user. What's wrong with my file ? Thanks for any help !

Serge Latouche : « La décroissance vise le travailler moins pour travailler mieux »

#UrgenceClimat. L’économiste de formation et théoricien de la décroissance prône d’entamer un sevrage à notre toxicodépendance à la consommation, passant par une rupture radicale, un changement de civilisation.

lemonde.fr/climat/article/2018

Afficher plus
Mastodon - etalab

Instance ouverte pour test - lire les CGU Elle est ouverte à tout agent possédant un compte email en ".gouv.fr" ou la liste disponible sur https://forum.etalab.gouv.fr/t/mastodon-le-reseau-social-libre-et-decentralise-en-plein-decollage/3538