🔥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

#TBT Our 1968 model Accelerator Operation Control center. It's a beaut! #SLACTurns60

Here's an ad for a postdoc at SLAC to work with on our project to use transformers / LLMs for theoretical particle physics (i.e. scattering amplitudes). academicjobsonline.org/ajo/jobs/23281

Wow, quite the story. 👇🏻👇🏻 Founding Arxiv is apparently not good for your academic career. Good evidence that improving how science works is insufficiently incentivized.

i think a lot about how people in academia shamed me for taking a summer off to do an internship in tech which paid 3x my yearly salary as a physics graduate student. they figured i was leaving for industry already or was not interested enough in my field of astrophysics.

Hamiltonian showing stable phase-space shrinking as the horizontal tune passes through the resonance

New research-y project: Blueprints for Intelligence, a visual history of artificial neural networks from 1943 to 2020 philippschmitt.com/blueprints-for…

One of the things I worked on during grad school around 2014-2017 was model predictive control for particle accelerators with learned models. In the mean time there's been slowly-developing interest in our field for RL. So this is interesting to see wrt things coming full circle.

I've been thinking about sustainability of scientific open source software more in the last few months. Something that has become clear to me is that having a maintainer team is not a side effect of a successful project, but (*effectively) a requirement for one.

Researchers develop clever #algorithm to improve our understanding of #particlebeams in accelerators @physrevlett doi.org/gr6xxn phys.org/news/2023-05-c…

Reviewer 1 “this paper does not contribute any new novelty or additional thought” Reviewer 2 “this paper is novel, convincing, and important” Sometimes I question why I even bother trying to publish in conferences. Without a discussion period what’s the point.

An “open science” approach to modeling particle accelerators could extend the capabilities of today’s machines and usher in the next generation, according to Axel Huebl & Jean-Luc Vay, researchers in the Accelerator Modeling Program LBNL Accelerator Technology & Applied Physics Berkeley Lab: atap.lbl.gov/open-science-e…

How do we know how smart AI systems are? | Science science.org/doi/10.1126/sc…

been spending most of my time on the “eh that was pretty obvious” end lately digging into the next “beginning to research” cycle

In the 1990s "double descent" was a well known property for even linear nets. As the number of train points = dim of the model (alpha=1) and noisy data the gen error spikes before decreasing (see dashed line). See tinyurl.com/yn2p85p6 for a full analysis in the linear case.

I am very happy about these recommendations from the #P5report. We deal with a lot of legacy infrastructure in accelerators. In order to fully benefit from the many advances in computing and AI/ML for science, we need to invest in infrastructure and software engineering talent.

The first two hours of this incredible eruption in Iceland in ONE MINUTE! Watch as the lava unzips along the fissure! #volcano #Iceland

"Perspectives on the State and Future of Deep Learning" (I really enjoyed taking part in this survey. Thanks to Micah Goldblum for organizing it!) arxiv.org/abs/2312.09323

Minnesota Sen. Gene Dornink (R-Brownsdale), during a hearing on sick time, calls a female pilot a “stewardess” and apologizes: PILOT: I’m a first officer for Delta. DORNINK: I’m sorry? PILOT: I’m a pilot.

Exciting work from our group showing how differentiable simulations can be combined with ML to provide reconstruction of the 6D position-momentum phase space of beams in minutes, with minimal data (tens of samples) and common accelerator equipment. journals.aps.org/prab/abstract/…

We often have models of accelerator behavior available that can be used as a starting point for machine learning based optimization. In this work our team explores one way system models can be used to speed up Bayesian optimization for accelerator tuning. nature.com/articles/s4159…