Arno Blaas

Autoregressive models dominate, but what if we treat multimodal generation as discrete order agnostic iterative refinement? Excited to share our systematic study on the design space of Tri-Modal Masked Diffusion Models (MDMs). We pre-trained the first Tri-Modal MDM from scratch on (text,), (image, text), and (audio, text). The same model can do ASR, TTS, T2I, captioning and native text generation. What I'm the most proud of in this work is the scientific rigor. Over 3,500 training runs. Principled hyperparameter transfer. Honest results. Carefully controlled ablations across multiple different axis of entanglement. A thread on our empirical findings (arXiV: arxiv.org/abs/2602.21472)

Our work on fine-grained control of LLMs and diffusion models via Activation Transport will be presented @iclr_conf as spotlight✨Check out our new blog post machinelearning.apple.com/research/trans…

Check out our MINT workshop on model interventions this Sunday!

Check out our latest work from our team at Apple MLR Barcelona!

Excited to announce that our workshop "I Can't Believe It's Not Better: Challenges in Applied Deep Learning" has been accepted at #ICLR2025! 🎉 Stay tuned for updates on the agenda, and paper submissions.

Do you work on foundation models? ➡️Submit to the @NeurIPS2024 🍃MINT 2024: Workshop on Foundation Model Interventions We feature a fantastic speaker lineup with @AtticusGeiger @hardmaru @JacobSteinhardt @viegasf, space for discussions, and more! sites.google.com/corp/view/mint…