Meta's SAM 3 and DINOv3 Models Enhance SYNAPS-I for Genesis Mission Projects

Meta's open-source AI models, SAM 3 and DINOv3, are being utilized in the SYNAPS-I initiative, a collaboration involving Lawrence Berkeley National Laboratory and three other national laboratories: Argonne, Brookhaven, and Oak Ridge. This initiative aims to analyze the tens of petabytes of data generated annually by the Department of Energy's light and neutron source facilities, equivalent to 2 million hours of HD video streaming.

The SYNAPS-I project employs 300 A100 GPUs to process data at a rate of 100,000 images per second. Researchers can expect a fully reconstructed 3D volume delivered in just 15 minutes, significantly speeding up the analysis process. Dario Gil, DOE Under Secretary, emphasized that the integration of AI with advanced computing and experimental systems allows for real-time data analysis and adaptive decision-making, replacing slower manual processes.

The Genesis Mission, set to launch in late 2025, aims to enhance scientific research capabilities through initiatives like SYNAPS-I, which currently involves 60 researchers. The deployment of these AI models at the National Energy Research Scientific Computing Center (NERSC) marks a significant advancement in the efficiency of data analysis in X-ray and neutron science.