Four Lawrence Berkeley National Laboratory scientists have received US Department of Energy Early Career Research Program awards, including funding for machine-learning work in scientific research and AI-guided control of major instruments. The awards put money behind projects aimed at making scientific models more reliable and accelerator systems more responsive.
Watch Desk analysis
What happened
A report by Mirage News says the four Berkeley Lab awardees will pursue projects spanning molecular science, scientific machine learning and accelerator control. Awards to DOE national laboratories are approximately $2.75 million over five years, according to the report.
Marcus Noack plans to develop scalable Gaussian-process methods that incorporate physics and calculate their own uncertainty. The work is intended to tackle false confidence in scientific machine-learning tools, run on DOE supercomputers and support reliable autonomous experiments through the open-source gpCAM platform.
Dan Wang’s project will develop physics-guided, hardware-aware AI and machine learning for real-time control of accelerators and lasers. The report describes planned demonstrations involving superconducting magnets, accelerator cavities, laser-beam stability and electron-beam shaping. Asmit Bhowmick will combine experimental structural biology, machine learning and modelling to predict water networks in biological systems and materials. Bingqing Cheng’s project will use physics-based simulations to study chemistry at electrified interfaces and defects.
Why it matters
These awards connect AI research to work where the constraints are not just accuracy on a benchmark. Models need to represent physical systems, communicate uncertainty and operate within the timing and memory limits of real scientific equipment. If the projects succeed, that could make machine learning more useful in experiments and instrument control, rather than leaving it as a clever answer on a screen.
The report describes planned research, not results already achieved. Wang’s proposed demonstrations are tests of the approach, not evidence that the systems have already delivered the stated benefits.
Our read
The most compelling thread is reliability: Noack’s models are meant to flag uncertainty, while Wang’s are designed around the hardware they must control. That is a better ambition than simply adding “AI-powered” to a scientific instrument and hoping nobody asks about the milliseconds.
The practical question is whether these projects produce methods other researchers can use, and whether the proposed demonstrations show dependable improvements under real operating conditions. The awards give the teams time and resources to find out.
What to watch
- Whether Noack’s methods improve uncertainty estimates while scaling on DOE supercomputers.
- What Wang’s four planned demonstrations show about real-time control and transfer to other instruments.
- Whether Bhowmick’s experimental data supports predictive models that work beyond the studied systems.
- What methods, software or results the projects publish during the five-year award period.
Discussion spark: For AI used in scientific instruments, should funders prioritise models that quantify their uncertainty, or systems that can act reliably in real time?
Sources and evidence
- Four Berkeley Lab Scientists Win DOE Early Career Awards – Mirage News (8 October 2026, 17:26 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.