AI Scientist @ Mistral prev: PhD student @ UCSD, Research Intern @ Google Research & MSR NE, Research Engineer at IBM Watson
San Francisco, CA
Joined June 2009
- Tweets115
- Following666
- Followers273
- Likes1.3K
Neil Mallinar retweeted
shoutout to @eaboix and @nmallinar for actually doing the work that made this paper happen, and Misha Belkin for advising.
paper here: arxiv.org/abs/2507.05644
Neil Mallinar retweeted
coauthors + I couldn't get a poster printed in Rio in time, so here's me ad-libbing at @iclr_conf
Neil Mallinar retweeted
1/ Deep learning is going to have a scientific theory. We can see the pieces starting to come together, and it's looking a lot like physics!
We're releasing a paper pulling together these emerging threads and giving them a name: learning mechanics.
🔨 arxiv.org/pdf/2604.21691 🔧
Neil Mallinar retweeted
Two generalization regimes in ICL: (1) context-scaling, where performance improves with more in-context examples, and (2) task-scaling, where performance improves with more pre-training tasks. While MLPs show task-scaling but not context-scaling,
arxiv.org/abs/2410.12783
Neil Mallinar retweeted
Iterating kernel ridgeless regression with AGOP computation groks modular arithmetic… and this grokking is remarkably similar to the phenomenon in neural networks.
I found these results very surprising!
Neil Mallinar retweeted
Say hello to our fellows attending/presenting at #ICML2024: Matt Levine, @skoularidou, @aldopacchiano, @WengongJin, @JiaqiZhangVic, @sana_tonekaboni, Ilker Demirel, @nmallinar, and Director Caroline Uhler. Good luck to Matt on his oral talk and Maria and Aldo on their workshops!
Neil Mallinar retweeted
Our workshop paper on Calibration is on arxiv -- check out @annabelle_cs's thread! I learnt a *lot* in this project -- ex:
1. ~All DNNs are calibrated *on their train sets*
2. ~Any model with small |TestError - TrainError| has good calibration (even if TestError is high!)
New ICML workshop paper 🚨arxiv.org/abs/2210.01964. Are deep neural nets calibrated? The literature is conflicted... bc the question itself has changed over time: as architectures, optimizers, and datasets evolve, it is difficult to disentangle factors which affect calibration.
Neil Mallinar retweeted
New ICML workshop paper 🚨arxiv.org/abs/2210.01964. Are deep neural nets calibrated? The literature is conflicted... bc the question itself has changed over time: as architectures, optimizers, and datasets evolve, it is difficult to disentangle factors which affect calibration.
Neil Mallinar retweeted
The paper is "The Calibration Generalization Gap", with @annabelle_cs @nmallinar @james_r_lucas. We're aiming to have it on arxiv "soon"... (bottleneck is me rn).
I can DM you the pdf in the meantime.
Neil Mallinar retweeted
Reminder that the ICML Workshop on Distribution-Free Uncertainty Quantification starts TOMORROW!
dfuq.rocks/22
We have 70 awesome papers on UQ for time-series, object detection, distribution shift, and more.
All are welcome in Room 308!
Neil Mallinar retweeted
Congratulations to @PreetumNakkiran for winning Harvard CS's Dissertation Award! If you want to read the award-winning thesis, it's on preetum.nakkiran.org/pdf/the…
Neil Mallinar retweeted
When you hear the news talking about "crime waves," think about what it means that the intentional arsenic poisoning of thousands of people in California prisons was never prosecuted. truthout.org/articles/this-p…
Neil Mallinar retweeted
My recent talk at the NSF town hall focused on the history of the AI winters, how the ML community became "anti-science," and whether the rejection of science will cause a winter for ML theory. I'll summarize these issues below...🧵
Neil Mallinar retweeted
"the gateway between dreams, trending on ArtStation" (CLIP guided YFCC diffusion, 968M parameter model, it's not done training yet)
Neil Mallinar retweeted
We have a talk on this work on Feb 11 at 9:30 am at AAAI. would love to chat at/after the talk.
Neil Mallinar retweeted
If you have faced the problems of generating the data for building chatbots, we present an iterative data programming method that helps data designers explore sparsely distributed examples faster. Excited to share our work that is accepted at IAAI/AAAI 20: ibm.biz/BdqFeV