🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

An interesting concrete suggestion for a mechanism to encourage voluntarily cooperation to reduce greenhouse gas emissions. If we can agree that climate change has a cost, then perhaps we can agree on a CO2 price, if only we had the right mechanism.

Late announcement, but I'm excited to announce that my PhD thesis is publicly available! In it I give my thoughts on how to approach variational inference in Bayesian neural nets, deep GPs, including discussion on the marginal likelihood and symmetries. arxiv.org/abs/2401.12418

Our paper, "The LLM Surgeon," accepted at ICLR 2024, achieves SOTA in LLM pruning in all unstructured, semi-structured, and the most challenging but most effective structured pruning that removes entire matrix rows/columns. Happy to share that code is now publicly available.

Build what you need and use what you build. This is a core philosophy of my research. It shifts the focus away from publishing “papers” to what really matters — impact. This thread unpacks why I think this is a successful approach to science. 1/10 Or see: perceiving-systems.blog/en/post/build-…

Great internship project by Tycho and his Qualcomm collaborators!

We have an opportunity for outstanding mid-career researchers (typically 10-20 years from PhD) to apply for a The Royal Society Faraday Discovery Fellowship at the department. Learn more here: cs.ox.ac.uk/news/2349-full… #compscioxford

A Gaussian Process regression model built on top of protein language model embeddings and inverse folding models makes accurate predictions with calibrated uncertainties. Wouter Boomsma

Looking forward to #GenU2024, my favourite gathering in the AI/ML calendar. The talks again all look incredibly interesting!

Test sets are the "get out of jail free" card of ML. It allows us to apply any crazy statistically-unprincipled method, and simply check if the predictions are good. But, it's easy to forget that holdout estimates themselves rely on assumptions that can be violated in practice!

Simultaneously funny and very relevant question… (Answer: no it doesn’t. But what’s missing, exactly?)

Within 2025, I will be hiring several PhD candidates and Postdoctoral researchers Imperial Computing to work on various aspects of #TopologicalDeepLearning (#TDL). Posts are attached. The postdoc positions are also available to apply under: imperial.ac.uk/jobs/search-jo… #CVPR #CVML #AI #ML

I'm recruiting a new PhD student on Gaussian Belief Propagation for Spatial AI. For this particular opportunity: 1. The funding I have is only available to a student who qualifies for "home fees" in England; 2. I'm looking for someone who can start soon, ideally April 2025.

I agree with Amanda Askell. It is a fallacy. It can be appropriate to say both a) that LLMs are next-token predictors, at a mechanistic level, and b) that they have understanding, at another level. (a) is an engineering fact. But the negation of (b) does not follow. 1/2

After using Deep Research for a while, I finally get the "it's just slop" complaint people have about AI art. Because I don't care much about art, most AI art seems pretty good to me. But information is something where I'm much closer to a connoisseur, and Deep Research is just

Understanding causes is key to science. Finite observational data alone isn't enough. While Bayes offers a framework to deal with this, the calculations are often intractable. We introduce a method to accurately approximate the posterior over causal structures. #ICLR2025 🧵1/15

Excited to be presenting this work at #ICLR2025. Please do reach out if you are interested in a similar space! 🗓️: Hall 3 + Hall 2B #471 🕐: Fri 25 Apr, 3 p.m. 📜: openreview.net/forum?id=eeJz7… This was a great collaboration w/ Matthew Ashman, James Requeima, Mark van der Wilk