The Happy Lab is hiring multiple PhD students for Fall 2025! RT appreciated :-) We discover how human factors can 1) mitigate ML-enabled abuse (deepfakes) and 2) be harnessed with ML-powered security to create safe, human-centric systems! Find us at: happyresearchlab.com :)

Interested in a #PhD that let‘s you transform what is possible with #AI & #MachineLearning? That gives you the chance to publish at the biggest conferences and offers unrivalled career prospects? Let’s talk! Join our Open Day and meet me and my students: vanderschaar-lab.com/join-the-van-d…

📢I’ll be admitting multiple PhD students this winter to Columbia University 🏙️ in the most exciting city in the world! If you are interested in dissecting modern deep learning systems to probe how they work, advancing AI safety, or automating data science, apply to my group.

Math 🤝 Halloween

We have already achieved self-improvement in diffusion models, via its own synthetic data! Not that far away definitely!

We've been rehearsing for over two months and we're ready to rock opening night tomorrow. If you're in Ottawa, don't miss out! You'll get to see *two* Castros for the price of one! 😅

Excited to be back to WiML at #NeurIPS2024 🥳

Interested in how LLMs are really used? We are starting a research project to find out! In collaboration w/ Sara Hooker Anka Reuel | @ankareuel.bsky.social Ahmet Üstün Niloofar (✈️ ICML) and others. We are looking for two junior researchers to join us. Apply by Dec 15th! forms.gle/H2o3cNCPdG8eDk…

I will be hiring 2~3 PhD students at Rice CS in the Fall of 2025 in the system for deep learning. Feel free to reach out and apply by Jan.1, 2025. No application fee! csweb.rice.edu/academics/grad…

no AI here, just the coolest paper i've seen in a while

I kinda miss when "AI Twitter" was folks doing AI research or app development, posting under their own identities. Slightly more intellectual depth than anime avatars whose main AI-related qualification is paying for a ChatGPT subscription.

Here's the thing. We have **NO IDEA** how to pick good graduate students. I served on admission committees for 10+ years, and chaired a few, and what I learned is that all the spreadsheets of grades and test scores and recommendations and essays and publications and interview

In 2021, when I decided to let go of deep learning theory and focus on improving math and reasoning in LLMs, I was inspired by the idea that one day LLMs could develop a solid theory explaining why deep learning works—and explain it to me in a way I can understand. Not too far

Checkout buddy Randall Balestriero discussing our work and more, why thinking of NNs as splines can provide strong intuitions of how they learn, generalize and grok! Also features a small cameo from yours truly!!

Indeed, it is that simple! The wiggliness induced by each layer allows NNs to approximate non-linear functions. More layers -> more possible wiggle -> more non-linearity. A nice way of thinking about this is imagining NNs doing origami on an elastic piece of paper!

My new paper "Deep Learning is Not So Mysterious or Different": arxiv.org/abs/2503.02113. Generalization behaviours in deep learning can be intuitively understood through a notion of soft inductive biases, and formally characterized with countable hypothesis bounds! 1/12

📣 The Journey Matters: Our #ICLR2025 paper shows how to pretrain sparse LLMs with half the size of dense LLMs while maintaining quality. We found that the average parameter count during sparse pre-training predicts quality, not final size. An MIT/Rice/Google/ISTA collab 🧵 1/N

Say goodbye to the silent era of video generation: Introducing Veo 3 — with native audio generation. 🗣️ Quality is up from Veo 2, and now you can add dialogue between characters, sound effects and background noise. Veo 3 is available now in the Google Gemini App for Google AI Ultra

GrokAlign: A method to accelerate grokking motivated by a geometric characterisation of the phenomenon. - The grokked state of a deep network is Jacobian aligned. - Centroids alignment provides an efficient and intuitive metric to monitor deep network training dynamics.

Introducing Genie 3, our state-of-the-art world model that generates interactive worlds from text, enabling real-time interaction at 24 fps with minutes-long consistency at 720p. 🧵👇