Fermilab scientists aim to create the next generation of physics tools to make new discoveries. To achieve this, they require novel tools that are more sensitive, more accurate and higher performing than those currently available. Artificial intelligence, or AI, is excellent at facilitating many of these tasks.
Particle physics experiments produce hundreds of terabytes of data per second. Because this large volume of data can’t be physically stored or transmitted elsewhere in its entirety, scientists must make rapid triggering decisions regarding whether to keep data for analysis or discard it. Fermilab scientists are using AI to optimize real-time selection algorithms, also called triggering algorithms, that can identify valuable data for analysis from the hundreds of terabytes generated per second by experiments. Their training models are learning how to distinguish processes of interest from irrelevant ones.
Scientists strive for fast and efficient trigger systems as well as those that produce valid results. Using so-called codesign techniques, Fermilab experts are seeking ways to combine AI with novel microelectronics to balance task and hardware constraints.
