Sylvain Marchienne

Photos Engineer at Apple. Apple Intelligence x Photos (Memory Creation). CS at @utcompiegne and @GeorgiaTech.

[Sylvain Marchienne @sylvainmarch]

@ElCuistoo J’ai eu la même ! Mais en déclassée, moins belle que toi, je l’embarque aux USA avec moi

[Ruoming Pang @ruomingpang]

As Apple Intelligence is rolling out to our beta users today, we are proud to present a technical report on our Foundation Language Models that power these features on devices and cloud: https://t.co/TaAdd0fBOp. 🧵

[Sylvain Marchienne @sylvainmarch]

@glazou Coke visuelle!

[Sylvain Marchienne @sylvainmarch]

@glazou @Apple @dashlane This is very true

[Daniel Glazman @glazou]

Shadow DOM, Web Components, WebExtensions and how @dashlane is leading on that front. With bits of yours truly inside. https://t.co/LEXIZ50B1g

[Sylvain Marchienne @sylvainmarch]

My team at @dashlane is looking for a Javascript Software Engineer (https://t.co/MK99Cvdvj4 ) and a Software Engineer Intern, Machine Learning (https://t.co/dTbmPYDAZp)

If you like low-level web development, ML engineering and research, this position may be for you!

[Sylvain Marchienne @sylvainmarch]

@ClementDelangue Les quais, je crois que ça fait l’unanimité de tous les runners Parisiens !

[Sylvain Marchienne @sylvainmarch]

@plhery @hexicans Mon iPhone 7 Plus de 2018 😬

[Sylvain Marchienne @sylvainmarch]

Copilot's model aka Codex from OpenAI: another huge language model trained on a huge dataset, without specific preprocessing. Impressive results for sure, but probably tricky to embed in real-time applications. https://t.co/hUiaRMH7fK

[Sylvain Marchienne @sylvainmarch]

TIL in this paper how encoding matrices can be used into Transformers: in the self-attention layer. Overall it interestingly demonstrates how possible it is to combine the two most trendy architectures at the moment: GNNs and Transformers!

[Tianle Cai @tianle_cai]

Transformer is a GOOD graph learner when incorporating suitable graph structural information!
Thrilled to share our new work:
Transformer + simple but effective encodings = GNN achieving new SOTA on OGB datasets, ZINC!
Paper: https://t.co/5r9mv6vRWq
Code: https://t.co/GhKbwUSMxs

[Sylvain Marchienne @sylvainmarch]

@jjerphan Très très bonne idée ça 👌⏹

[Sylvain Marchienne @sylvainmarch]

2020

[Daniel Glazman @glazou]

Via une collègue qui se reconnaîtra. J'ai explosé de rire :-)

[Sylvain Marchienne @sylvainmarch]

Impressive robot capabilities from @pollenrobotics ! Congrats les Bordelais 🙂 https://t.co/9cVSmAqIHD

[Sylvain Marchienne @sylvainmarch]

I always ***dumbly*** think of docker-containerizing my Python models as the final, simplest, and fastest step of my job. And every time, I tell myself "at least I learned something today". Sure.

[Sylvain Marchienne @sylvainmarch]

"Que vient faire le machine learning dans un logiciel de mot de passe ?" Enfin une explication claire, bravo @glazou ! @dashlane (2:00-10:00) https://t.co/stjDXOgqIr

[Sylvain Marchienne @sylvainmarch]

Come and try! https://t.co/0X2t6KcMnd

[Sylvain Marchienne @sylvainmarch]

@MrAstroThomas Exciting! Let me know if you get satisfying results with this method

[Sylvain Marchienne @sylvainmarch]

DBSCAN is a great algorithm for geo clustering. Using the Haversine distance and lat/lon coordinates, it yields consistent clusters. In this article, I introduce a way to integrate more features into the geo clustering process.
https://t.co/dOpwEeTCBs

[Sylvain Marchienne @sylvainmarch]

@GygliMichael I have been reading a lot about Video Summarization recently and it seems to me me that you pioneered the task. To your mind, what's the current state of the art method? Supervised? Unsupervised?