Postdoctoral Research Scientist Position at Columbia U.

Job Description:

Nevis Laboratories at Columbia University has an immediate opening for a Postdoctoral Research Scientist to work on the research and development of physics-informed, real-time AI data-processing pipelines, for high-rate streaming data. Successful participation in these projects will involve collaboration with the Neutrinos and Rare Events group at Columbia University, and research partners (including engineers, computer scientists and physicists) at US universities and national labs. The successful candidate will be expected to dedicate substantial effort into the development and optimization of real-time AI data processing pipelines in custom-designed, mixed hardware architectures (CPU/GPU/FPGA). The work will target applications for large-scale, high-resolution particle imaging detector projects in the US, with opportunities for journal publications and presentations at national or international meetings and conferences. For more details, see below.

The appointment is for a one-year term with the potential of an extension on a yearly basis to up to three years, contingent upon satisfactory performance and available funding. The position can be based at Nevis Laboratories or Morningside Campus at Columbia University.

Review of applications will begin at the time of receipt of required materials, and will continue until the position is filled. To ensure full consideration, candidates should apply by submitting a cover letter, CV, publication list, and brief statement of research interests and experience by February 28, 2023. In addition, they should arrange for three letters of recommendation to be submitted on their behalf by the same deadline. All application materials should be submitted to

Job Requirements:

Applicants should have a strong interest in software/algorithm development, particularly adapting AI algorithms for implementation in commercial or custom hardware accelerators. Applicants should hold either a recent PhD in physics (preferably experimental particle or nuclear physics or in a related field), or a recent PhD in computer science (preferably in hardware systems/computer architecture or a related area). Background/training in hardware system-level design and FPGA-based acceleration; proficient use of High Level Synthesis; experience with machine learning algorithms and inference; and interest in machine learning inference in FPGAs are strongly preferred.

At Columbia University we value diversity as a way to enrich our scientific research and academic missions. We strongly encourage applications from members of groups who are underrepresented in STEM. Columbia University is an Equal Opportunity/Affirmative Action employer — Race/Gender/Disability/Veteran.

For any questions or further information on these positions, please contact: Prof. Georgia Karagiorgi

Letters of Reference and application materials should be sent to

Additional Details:

Real-time AI for High-rate Streaming Visual Data

Modern particle detectors, for example liquid argon time projection chambers, work by “imaging” particles and their trajectories with high repetition rate and high resolution. The resulting high data rates (often up to multiple terabytes per second) require fast, efficient, and intelligent processing of the data, and in a parallelized way, in order to select rare and/or faint signals of interest with high accuracy while minimizing background and noise contamination as much as possible. As part of this project, you will be joining a team of particle physicists and computer scientists who are interested in adapting AI algorithms for implementation in commercial off-the-shelf or custom-designed hardware accelerators for this purpose. This is a targeted application for the planned Deep Underground Neutrino Experiment (DUNE), a major international particle physics project which will be hosted in the U.S., but also lends itself to other particle physics or astro-particle physics experiment applications, and beyond. Collaboration on these projects entails AI algorithm optimization for the purposes of reducing its computational footprint without compromising accuracy using techniques such as pruning or quantization-aware training, and subsequently implementation and testing of such algorithms in FPGAs, using HLS tools, e.g. hls4ml.

Related publications:

Yeon-jae Jwa, Giuseppe Di Guglielmo, Lukas Arnold, Luca Carloni, and Georgia Karagiorgi, Real-Time Inference With 2D Convolutional Neural Networks on Field Programmable Gate Arrays for High-Rate Particle Imaging Detectors, Front.Artif.Intell. 5 (2022), 855184. e-Print: 2201.05638 [physics.ins-det]. DOI: 10.3389/frai.2022.855184