Reminder that, if you're applying to grad school, postdocs, faculty positions etc., you should have asked your writers 2 to 4 weeks ago. The second best time to ask is today. Here's my post on how to ask for a letter of recommendation. kamathematics.wordpress.com/2021/08/18/how… 1/2

Deadline extended for DAI workshop at AAAI 2025 26 November 2024 AoE. IIT Madras @

The 3rd Workshop on Deployable AI (DAI) will be held in conjunction with AAAI 2025. Submissions due: November 26, 2024 (AOE) *Extended Deadline* Website: lnkd.in/dfXVP9zS Submission Link: lnkd.in/dWs97Sg6 important technical challenges and ABOUT: Artificial

I'm looking for PhD applicants who have expertise in Gaussian processes and/or Transformers for an exciting PhD project If this sounds interesting, application deadline for funding is 3/12 Please share with people you think this might be relevant to! oatml.cs.ox.ac.uk/apply.html

🚨 PhD position alert! 🚨 I'm hiring a fully funded PhD student to work on mechanistic interpretability at UvA Amsterdam. If you're interested in reverse engineering modern deep learning architectures, please apply: vacatures.uva.nl/UvA/job/PhD-Po…

Lecture Notes on High Dimensional Linear Regression ift.tt/QoBcJUK

#AAAI2025 📢 Deployable AI AAAI ⭐ Deployable AI Workshop 🗓️ March 3, 2025 📍Room 120A, Pennsylvania Convention Center 🔗sites.google.com/view/dai-2025/… Balaraman Ravindran @rahulvash Danish Pruthi Aditi Raghunathan Krishna Pillutla RBC-DSAI@IITM #AI #ML #LLMs #AIEthics #AIForSocialImpact

The 3rd Deployable AI workshop AAAI 2025 is tomorrow 3rd March 2025, starting at 9 AM Room 120A, Pennsylvania Convention Center, Philadelphia Details: sites.google.com/view/dai-2025/… Balaraman Ravindran Arun Rajkumar Gokul S Krishnan Eugene Bagdasarian Vishal Misra Arpita Biswas Harish Ramaswamy

EPFL's "Optimization for Data Science" PDF: n.ethz.ch/~jiaxie/gradua…

Thanks to all the participants and invited speakers for making the third edition of the DAI Workshop a success at AAAI. Insightful keynotes & an engaging panel discussion. Accepted works: sites.google.com/view/dai-2025/… Balaraman Ravindran Eugene Bagdasarian Vishal Misra Gokul S Krishnan Arun Rajkumar

You'd think how wounds heal would be well understood. But you'd be wrong. Tannie Liverpool explains how medics work with mathematicians & other scientists to apply more than just a sticking plaster. Oxford Mathematics Public Lecture, online, 5pm, Wed 12 March: youtu.be/i-nAaIDJu0c

An Imperial College eng professor gave four LLMs a problem set that graduate students had two months to solve. He had TAs grade the results blind alongside real submissions. Meta AI and Claude failed. ChatGPT ranked 27 of 36 students...while Gemini 2.5 Pro ranked 4 of 36 🤯

As we go through a lot of excitement about RL recently with lots of cool work/results, here is a reminder that RL with a reverse KL-regularizer to the base model cannot learn new skills that were not already present in the base model. It can only amplify the existing weak skills.

The greatest feeling you can have as a scientist is realizing you’ve discovered something that no one else in the world knows yet When this happens, put your pencil down, take a deep breath and savor those three minutes before someone points out it was published fifty yrs ago

Why do gradients increase near the end of training? Read the paper to find out! We also propose a simple fix to AdamW that keeps gradient norms better behaved throughout training. arxiv.org/abs/2506.02285

Emergence in transformers is a real phenomenon! Behaviors and capabilities can appear in models in sudden ways. Emergence is not always just a "mirage". Compiling some examples here (please share any I missed): 🧵

Here's an open problem in optimization, which is easy to state but is extremely challenging: Nesterov's acceleration requires strong convexity μ and smoothness L. There exist variants that estimate L adaptively. Can we estimate μ if we know L? Partially solved in this new paper,

You may think of optimization and linear algebra as established fields, but consider this: until 2008, convergence rate of SGD applied to linear systems wasn't established. I'm sure there are many other discoveries waiting to be made, things we'd be surprised later weren't known.

1/ 🚨 New paper alert! 🚨 We explore a key question in deep learning: Can independently trained Transformers be linearly connected in weight space — without a loss barrier? Yes — if you uncover their rich symmetries. 📄 arXiv: arxiv.org/abs/2506.22712

The best logical explanation of entropy. (It only took me 11 months to find an hour to read it). arxiv.org/abs/2409.09232