ML researcher @ Universal Music Group | Prev. PhD @ Sony CSL × Télécom Paris | Very interested in AI, music and AI for music
Joined July 2022
- Tweets395
- Following369
- Followers290
- Likes2K
Pinned Tweet
❌ We don’t need no negative samples
❌ We don’t need no large batches
❌ No modality gap in the classroom
Very happy to introduce SLAP, our latest brick in the wall of multimodal SSL 🎶🧠
Joint work with King @Juj_guinot, accepted at #ISMIR2025! 🇰🇷
1/7
Alain Riou retweeted
Our paper: Equivariant Music Transformer has been accepted to ISMIR 2026 (Best Paper Award Nominee). We show that music transformers become less equivariant as they scale or train longer, suggesting they memorize each musically-shifted version of the same data separately. 🧵👇
Alain Riou retweeted
1/12 Introducing Music Lenses!
Something cool you can do with our recent sparse-inversion work is prompt a dense music space for concept-conditional similarity. You want similar vocal character, rather than just similar songs? Well, now you can!
pliploop.github.io/sparse-st… 🧵
Happy to share our #ISMIR2026 paper on steerable music retrieval with sparse autoencoders 🎛️🎧
Dense retrieval is great at finding similar tracks,
But once you have a query, what if you want to modify one concept inside it?
More piano,
More intimate,
Less gritty.
1/n
Alain Riou retweeted
Information exclusive Blast : le Premier ministre a préparé la coupure des réseaux sociaux pour le 24 octobre
Deux jours avant la manifestation du 24 octobre, le Premier ministre Sébastien Lecornu a demandé à Orange, SFR, Free et Bouygues Telecom de se tenir prêts à couper l’accès à TikTok, Instagram ,Snapchat et X. Blast révèle la lettre et le compte rendu de la réunion qui a suivi, où les opérateurs ont exprimé leurs doutes. Le gouvernement avait déjà utilisée cette mesure liberticide en Nouvelle-Calédonie en 2024, avant d’être désavoué par le Conseil d’État.
Par @yanmdi.
blast-info.fr/articles/2026/…
Alain Riou retweeted
"A Fix for Softmax and Cross-Entropy"
Turns out two of the most fundamental functions in deep learning have numerical bugs that can actually destabilize training.
This paper proposes simple fixes to softmax and cross-entropy that prevent tiny gradients from incorrectly rounding to zero, or remaining nonzero when they should vanish, as these errors can cause sudden loss spikes and interfere with knowledge distillation.
On small-scale grokking benchmarks, the fixes eliminate Slingshot loss spikes and improve test accuracy from 94.3% to 100% in one setting. They also prevent student models from drifting away from their teachers during distillation, without requiring higher numerical precision.
alphaxiv.org/abs/2610.31642
Alain Riou retweeted
1/ JEPA selects which features its world model keeps.
But how?
Will it keep a feature it can't predict, like a traffic light?
Judging each feature alone, earlier work says no.
We show evidence that this is wrong and introduce a new characterization that explains why. 👇🧵
Alain Riou retweeted
This semester I’m teaching "Fundamentals of LLMs" at the MVA master in Paris 🎓
I’ve turned my 1st lecture, "NLP before LLMs", into a blog post that goes from simple word counts all the way to attention.
The approach is learning by doing 💪 : each step comes with hands-on exercises inspired by key research papers.
Why practice by hand when an AI agent can write the code for you ? Precisely because it can ! When code becomes easy and cheap, understanding is what matters, and you will build it by doing the work yourself.
Have fun learning by doing !
👉 huggingface.co/spaces/dlouap…
(Next up: Tokenizers. Stay tuned !)
What is an intuitive way to search in a music library? Tags? Captions? Chatbots?
We propose another paradigm: what if we could create a slider for *any* musical concept and use it to refine in real-time what we're looking for? 🎛
Work led by @Juj_Guinot, accepted at @ISMIRConf!
Happy to share our #ISMIR2026 paper on steerable music retrieval with sparse autoencoders 🎛️🎧
Dense retrieval is great at finding similar tracks,
But once you have a query, what if you want to modify one concept inside it?
More piano,
More intimate,
Less gritty.
1/n
Alain Riou retweeted
In France, even the police loots
‼️🔴‼️
C’est la crise pour tout le monde même pour la police 👌
‼️Une autre vidéo ou on voit des policiers dérober des objets en toute impunité
Hésitez pas à partager
———
#17octobre
Alain Riou retweeted
Voir des politiques de droite qui ont arrêté les maths en seconde utiliser le classement PISA pour décrédibiliser les lycéens c’est le summum du foutage de gueule venant de gens incapables de calculer un pourcentage
Alain Riou retweeted
La bourgeoisie aime la révolte... Au programme du bac, dans les musées ou au Panthéon. Vivante, elle préfère lui crever un œil. ⤵️
Alain Riou retweeted
DINOv2 has never seen a caption, and Qwen3 has never seen an image. We still aligned their embedding spaces without a single image-caption pair.
It even works when the images and the captions come from different datasets.
Project page: dominik-schnaus.github.io/un…
⬇️
Alain Riou retweeted
Les images des manifestations qui touchent la France ont été adaptées en manga et font des millions de vues sur les réseaux.
Attendez de voir la suite...😨
Crédit : Madara9314 sur TikTok
Alain Riou retweeted
To be presented at ISMIR 2026 ☮️ PEACE, a joint embedding of audio effects programs written in Faust and their corresponding "wet"/"processed" audio.
It enables audio-to-code retrieval when given a library of programs and their embeddings.
Joint embeddings have applications beyond retrieval, and the paper couldn't address all of them experimentally. Think about CLIP/CLAP and make the analogy to rich understanding of digital signal processing code (DSP) and music.
Read the paper: arxiv.org/abs/2610.03405
and codebase: github.com/DBraun/PEACE
More findings from the paper 🧵
Alain Riou retweeted
Nunez doit démissionner. Quelles que soient les circonstances, on n'envoie pas des grenades sur des enfants. Cette stratégie du pourrissement a déjà coûté trop de mutilations d'enfants.
Mais comme d'habitude plutôt que de démissioner on aura droit à un "j'assume".
This quoted post is unavailable.
Alain Riou retweeted
SFT is not dead! 🥳
We found a way to make SFT rival current prevailing posttraining methods, often generalizing better and forgetting less than RL and OPSD. 🤯
Following our prior work on reasoning with sampling, we now introduce sampling to the posttraining stack.
1/n
Alain Riou retweeted
Ben Hayes, Haokun Tian, Stefan Lattner, "Synth-JEPA: Joint Embedding Prediction for Renderer-Free Synthesizer Parameter Search" arxiv.org/abs/2609.31024
Alain Riou retweeted
Ben Hayes, "Off-manifold robustness in synthesizer inversion with joint distribution flow matching" arxiv.org/abs/2609.29320
Alain Riou retweeted
Accepted at #NeurIPS26. See you in Paris (hopefully).
