Edge artificial intelligence for scientific instrumentation
par
IJCLAB
Radiation and particle detectors let us observe the universe beyond what our senses can achieve. Scientists come up with new ways to collect information about our universe using detectors to transform particles into electrical signals and digital bits. These novel detectors are more precise, larger and offer much more range than their predecessors, resulting in large amounts of channels (millions) and data (>100 TB/s). Transmitting this raw data continuously to a compute system requires mechanical space, energy and a significant amount hardware, increasing costs and making some experiments non-viable.
My research group develops systems to distribute real-time compression and analysis algorithms along the data acquisition chain including right up at the edge of the detector, reducing data at or near the source. However, this approach presents several challenges with limited computing power and memory availability. We combine small AI models with carefully selected signal processing and compression algorithms to achieve very efficient smart acquisition systems capable of several orders of magnitude data reduction in a few microseconds.
GDR DI2I