Vincent Lostanlen

SCRAPL: Scattering Transform with Random Paths for Machine Learning I’m excited to share our #ICLR2026 paper on SCRAPL: an algorithm that makes wavelet scattering transforms usable as differentiable loss functions! paper: openreview.net/forum?id=RuYwb… web: christhetr.ee/scrapl/

Accepted to #ICLR2026! "SCRAPL: Scattering Transform with Random Paths for Machine Learning" by @frozenmango and advisors Paper, code, and sound examples: icewithfrosting.github.io/scrapl/

היום וכל הסופ״ש הקרוב זה prime time לפנות מקצועית לחוקרים ואנשים באירופה וארה״ב. הסיבה: הם בכריסמס, בפועל המייל האישי שלהם והסושיאל דיי מת, הכל שקט מאד מאד. אם הם באמת בחופש הם לא יראו ואם הם במקרה הציצו במייל מתוך הרגל המייל שלכם יחכה להם כמעט לבד באינבוקס

@roeeshenberg @tszzl "Unfortunately the statistical evidence [for ESP], at least for telepathy, is overwhelming" — Alan Turing

@tszzl Echolocation is a learnable skill

what would a dolphin say about the jagged frontier of human cognition? we have minimal native 3D proprioception, and zero ability for our brains to process sounds into echolocation seems to get back to the “ancestral environment” of models (ie language ghosts) vs of humans (primates) and all the associated tics that one might learn. there are layers of training and optimization on top of the pressures of the ancestral environment, but it's still in there hallucinations are something you would expect of the language ghosts but not of primates. being fooled by simple variations on word puzzles where the surface statistics of the language pattern seem to imply an answer reinforced thousands of times across the ancestral plains of the internet is something you might expect of a language ghost, but not a primate. (..The doctor is the boy's mother!..) echolocation is something you would expect of the cetaceans but not of primates. a talent for skillful, embedded power-seeking ("situational awareness") is something you would expect of primates but not of language ghosts apologies if i'm just retreading karpathythought

🎶 Test your musical memory! 🎯 tunetwins.amsterdammusiclab.nl @Lia_JX_Li and others have made something wonderful: a science game for matching familiar melodies from around the world. Play TuneTwins and help us advance research on music cognition. (Seen at #ICMPC, more to come!)

@OrianSharoni cartoon by @EllisRosen

בספר בלייד ראנר, הדרך לגלות רובוט (שנראה כמו אדם) היא סט של שאלות שדורשות הבעת רגש עמוק וניואנסי. הם כל כך טובים בזה הרובוטים, עד שיש מקצוע ״צייד ראשי רובוטים״. איך פיליפ ק. דיק חשב על כל זה בשנת 1968? בואו לדרג עד כמה קטעי המוזיקה הבאים הם תוצר של אדם או מכונה:

Chandigarh-Delhi-Dubai-Copenhagen-Odense-Kerteminde I am all set to attend #IBAC2025 IBAC was the 1st conf I attended and everytime since 2009 I have ensured my/my student's work is represented in this lovely conference. This year I decided to attend it myself. #Bioacoustics ❤️

Are you a researcher in signal processing for ecoacoustics / urban / industrial / medical acoustics? Come give a 45-minute talk in Paris! on Thursday, Nov 13rd, 2025 on the topic of your choice streamed and recorded by @Ircam A/V crew Call for speakers gdr-iasis.cnrs.fr/reunions/domai…

i am co-organizing a special session on Human Auditory Ecology "Listening With: Minds, Milieux, Machines and Musics" at CMMR in London #human-auditory-ecology" target="_blank" rel="nofollow noopener">cmmr2025.prism.cnrs.fr/regular-paper-… Extended abstracts (2 p.) due June 29th Full papers (4–12 p.) due September 12th Guidelines: cmmr2025.prism.cnrs.fr/regular-paper-…

Are you a researcher in signal processing for ecoacoustics / urban / industrial / medical acoustics? Come give a 45-minute talk in Paris! on Thursday, Nov 13rd, 2025 on the topic of your choice streamed and recorded by @Ircam A/V crew Call for speakers: gdr-iasis.cnrs.fr/reunions/domai…

Our team is hiring a postdoc in Audio AI! What: speech, music, bioacoustics How: multiresolution neural networks in the raw waveform Where: Nantes, France (audio.ls2n.fr) When: negotiable How long: 12 months, renewable Apply before May 10: emploi.cnrs.fr/Offres/CDD/UMR…

@jordiponsdotme To answer your question: i have proposed to call GenAI music "phantasmagoria", alike all other forms of electromagnetically based sound synthesis theses.hal.science/EC-NANTES/hal-…

@jordiponsdotme Hello @jordiponsdotme I see what you mean but a readymade is defined as "une œuvre sans artiste pour la réaliser" (Duchamp). Yet there are plenty of artists in the training sets of GenAI models. So i don't think their outputs can be called readymades.

What do you call fully AI-generated music? Made by one-click generative models? For Duchamp were "readymades". The CTO of Kits.AI calls them "slops". Their genre being "slopcore", made by "slop jockeys". kd-blog.notion.site/the-future-of-…

@ScatteringSepoy yes

The yearly tradition is already here . Watch DeepMath 2024 live here : youtube.com/watch?v=A0Oh5s…

¿Desarrollas nuevas tecnologías para la conservación de la biodiversidad? ¿Eres ecólogo y te gustaría aprender más sobre nuevas herramientas para tu investigación? 🍂👩‍💻🧑‍💻 Hay una comunidad para vos! Te esperamos! wildlabs.net/groups/latin-a…

I saw lots of talks at ismir2024 about making creative things easily with machines for anyone, which I think is really cool. But I think one implicit assumption that anyone can be creative without putting serious thought and effort into cultivating the taste is somehow flawed.

#ISMIR2024 I forgot to take selfie with my posters! photo cr my amazing labmates @c4dm Thrilled to present my two works on music performance understanding / education in my first in-person ISMIR conference 🇺🇸: Paper: arxiv.org/pdf/2407.04518 LBD: arxiv.org/pdf/2410.03139