Kymatio

Open-source Python library for wavelet scattering transforms.

[MLIA @mlia_isir]

D-day for Edouard Oyallon who is defending his HDR entitled "Contributions to Local, Asynchronous and Decentralized Learning, and to Geometric Deep Learning"

[Lia Jiaxin Li @Lia_JX_Li]

The Music Cognition Group is excited to welcome Dr. @lostanlen for a captivating talk on music signal processing next week @UvA_Amsterdam. Really glad our connection from Timbre 2023 has evolved into collaboration! Check out the poster for more details! #musicscience

[Vincent Lostanlen @lostanlen]

#ISMIR2023 final credits

Gotta hand it to @cyrusasfa and @frozenmango for turning ten years of research on time–frequency scattering into a series of Colab notebooks that anyone can use.

Thanks to all the attendees. I was very pleased with the turnout, up until 6pm!

[Christopher Mitcheltree @frozenmango]

If you're attending ISMIR and curious about wavelet transforms applied to audio, come check out our tutorial on Sunday afternoon! 🌊🌊🌊

[Vincent Lostanlen @lostanlen]

"Mesostructures: Beyond Spectrogram Loss in Differentiable Time–Frequency Analysis" is out!

with @cyrusasfa, Han Han, @Tsingmoon, Mathieu Lagrange, and Gÿorgy Fazekas

made with @KymatioWavelets

[Kymatio @KymatioWavelets]

😏 https://t.co/m27h0cKSLP 🌊

[François Chollet @fchollet]

What if all you need is nearest-neighbors + a really good compression algorithm 🤔

[Vincent Lostanlen @lostanlen]

PhD for hire!
"Theory and implementation of multi-resolution neural networks"
with applications to #audio #AI

at @LaboLS2N (Nantes🦀)
in partnership with @ari_oeaw (Vienna 🎻)

duration: 3 years
funding: @AgenceRecherche
start date: negotiable

https://t.co/zD8iei5ICY

[Kymatio @KymatioWavelets]

At #ISMIR2023? Join our tutorial
"Deep Learning meets Wavelet Theory for Music Signal Processing"
by @cyrusasfa, @lostanlen, and Ch. Micheltree

https://t.co/ormumF7v5w

On Sunday, Nov. 5, afternoon 🌊

[Kymatio @KymatioWavelets]

@BillBainbridge5 @cyrus_vahidi @dafx2020 @lostanlen @Tsingmoon @c4dm @CDT_AI_Music @LaboLS2N That's right :)

Here's an example of a 12'' record made with JTFS

Lorenzo Senni "XAllegroX" remixed by Florian Hecker
https://t.co/MyL6qZswF6
out now on @WarpRecords

[Kymatio @KymatioWavelets]

Two papers from @LaboLS2N at @CentraleNantes presented today at two different events!

Eco-acoustics with Scattering1D and k-determinantal point processes 🐦🦗🦜
https://t.co/Ja9Fa8Ssdl

Differentiable time-frequency scattering on GPU 🎤🎛️🔊
https://t.co/cHIHmIaRiq

[Vincent Lostanlen @lostanlen]

Today in @KymatioWavelets news

* Mohamed Outidrarine and Pierre Baudet present k-determinantal point processes for eco-acoustic exploration at #gretsi22 (Nancy, FR)

* Han Han, @cyrus_vahidi, and @Tsingmoon get best paper award for differentiable JTFS at #dafx22 (Vienna, AT)

[Kymatio @KymatioWavelets]

@meickenberg @lostanlen @cyrus_vahidi @Tsingmoon @ScatteringSepoy @PhilipWarrick Joint Time Frequency Scattering is just around the corner 👀

https://t.co/iQZZQpB9qQ

[Kymatio @KymatioWavelets]

Kymatio 0.3.0 "Edre" has been released!

New features include a brand new Jax backend, customizable Q₂ in 1D, customizable averaging, refactoring of the 1D filter generation, improved examples, bugfixes, and much more!

Check out the update at https://t.co/dUkq5tFFky!

[Kymatio @KymatioWavelets]

Thanks to all our contributors!

@meickenberg @lostanlen @cyrus_vahidi @Tsingmoon @ScatteringSepoy @PhilipWarrick and many more not on Twitter

[Vincent Lostanlen @lostanlen]

I am looking for a PhD student in MIR! 🎓
to work on self-supervised learning in multitrack music

You will be employed by @Deezer Research
co-advised by R. Hennequin + me
based in Paris 🪗 and visiting my lab in Nantes 🦀

Start: Dec. 2022☃️ or later
https://t.co/cvGIFCzq7q

[Kymatio @KymatioWavelets]

@ScatteringSepoy Invented by @StephaneMallat and developed by notables such as @joanbruna, @laurentsifre,Irène Waldspurger... and the @KymatioWavelets team 🎉

[Kymatio @KymatioWavelets]

The hope is that this machine would run only our benchmarks when scheduled and that hardware changes would be few to zero over large scales of time.

Having access to this would accelerate Kymatio development, which would accelerate research in signal processing. 4/4

[Kymatio @KymatioWavelets]

Kymatio is lookin for a machine with GPU to run benchmarks on!

If you have a lead on this request (or alternative considerations) please DM us or email the following email address;
muawiz DOT chaudhary AT mila DOT quebec

(Retweeting also helps!) 1/4

[Kymatio @KymatioWavelets]

Benchmarks are then vital to know the runtime of our implementations, or how much our changes and cool new abstractions impact speed (if any).

Benchmarks could help to prioritize certain changes, identify possible slowdown, or give maintainers peace of mind. 3/4