Kevin Scaman

#Article 📰 | Comment repérer des communautés sur les #RS ? Comment garder le contrôle de ses données sur le #web ? Des questions auxquelles la nouvelle équipe ARGO (@Inria/@ENS_ULM) tente de répondre en alliant #algorithmique des #graphes et #ML. Lire 👉 inria.fr/fr/allier-math…

The Chomsky et al. opinion piece in the @nytimes about ChatGPT is making the rounds. Rather than trying to deconstruct their argument, I asked @bing what it thinks of it. Now you can judge for yourself who has the moral high ground 😂.

𝗣𝗿𝗼𝗯𝗹𝗲𝗺𝘀 𝘄𝗶𝘁𝗵 𝗴𝗿𝗮𝗱𝗶𝗲𝗻𝘁 𝗱𝗲𝘀𝗰𝗲𝗻𝘁 𝗮𝗻𝗱 𝘀𝘆𝗺𝗺𝗲𝘁𝗿𝘆 If the loss has symmetry, several solutions are equally good, and to converge to the 'right' solution, gradient descent needs additional information. 👇 a simple example below 👇

𝗡𝗮𝗺𝗲𝗱 𝗧𝗲𝗻𝘀𝗼𝗿𝘀 allow users to give explicit names to tensor dimensions. Named Tensor Notation by @davidweichiang @srush_nlpand @boazbaraktcs are super convenient. Here is a simple transformer encoder block! 👉tiny.cc/l0x2vz

Our workshop on Symmetry and Geometry in Neural Representations has been accepted to @NeurIPSConf 2022! We've put together a lineup of incredible speakers and panelists from 🧠 neuroscience, 🤖 geometric deep learning, and 🌐 geometric statistics.

Today at 8 pm (CEST), I am presenting our work with @giannis_nikole, @KevinScaman, @Monoideabelien, @mvazirg on "Ego-based entropy measures for structural representations on graphs" to #ICASSP2021! Tune in here: bit.ly/3v8dXyk Proceedings: ieeexplore.ieee.org/document/94142…

Understanding convolutional neural networks from first principles. A great opportunity to learn about circulant matrices, shift-equivariant layers, discrete Fourier transform, CNNs, receptive field... and play with @JuliaLanguage All the details here👇 dataflowr.github.io/website/module…

Our work "Lipschitz Normalization for Self-Attention Layers with Application to Graph Neural Networks" has been accepted as a short talk to #ICML2021! Thanks @Monoideabelien @KevinScaman for the great collaboration! Check out our paper for more details! arxiv.org/pdf/2103.04886…

Please join us tomorrow at 7AM CET for presentations of the 11 @Huawei papers at #NeurIPS2020. The last two papers are from my team @HuaweiFR. @KevinScaman @LDosSantos_ @lgorcolin @merwanbarlier Cedric Malherbe us02web.zoom.us/j/5797113087?p… Conference ID: 579 711 3087 Password: 790071

Choosing the best algorithm to solve an optimization problem IN PRACTICE depends a lot on the data/parameters/implementation/etc. An objective selection needs time consuming benchmarks. This can now be done easily and together on GitHub with BenchOpt! benchopt.github.io

If you like Gaussian kernels and distributions, you will enjoy this month blog post on Hermite polynomials! francisbach.com/hermite-polyno…

Please to announce that our paper "A Simple and Efficient Smoothing Method for Accelerated Optimization and Local Exploration" has been accepted to #NeurIPS2020 🥳 Co-authors @lgorcolin @KevinScaman @MerwanBarlier

#IJCAI proceedings are online. Check out our paper "Coloring Graph Neural Networks for Node Disambiguation" introducing an universal approximator of continuous functions on graphs. Paper ijcai.org/Proceedings/20… With George Dasoulas @Monoideabelien @KevinScaman