Speaker
Description
There is a growing demand for intelligent instrumentation to enable rich data extraction from detectors without overwhelming data rates. Machine Learning (ML) deployed close to the detector, in the data acquisition chain, provides opportunities to select and efficiently compress relevant data. In 2024 both the CMS and ATLAS experiments at the LHC have used ML in their first stage hardware event filters to select interesting collision data from the sea of background. This achievement showcases the strong cooperation between ML experts, domain scientists, and hardware developers in the adoption of ML ‘at the edge’. Techniques and tools enabling the efficient use of ML in hardware include extreme quantization; optimized ML architectures; and low latency, low power, and high throughput hardware implementations. This talk will describe the technology behind edge ML, as well as applications in particle detectors and other domains including medical imaging and earth observation, demonstrating the transformative potential across diverse fields.
