Low latency Edge AI for Physics searches at 40 MHz

Despite having analysed of over 300 inverse femtobarns of data from the CMS experiment at the LHC, no evidence for New Physics beyond the Standard Model (SM) has been found. Yet, the SM is understood to be an effective theory, and a more fundamental theory is expected at smaller scales. A major limitation of current searches is that only 0.02% of the total data produced by CMS is used for analysis, while nearly 99% of the data is discarded within microseconds by the trigger system, before it even leaves the underground experimental cavern. This discarded data, filtered based on known physics signatures, may contain exactly the kind of unconventional signals that current analyses miss. The aim of this project is to dramatically increase the discovery potential of CMS by developing an online system that can analyse every proton collision event at the full 40 MHz collision rate in a signal-agnostic way. By targeting the 99% of lost data, this project opens new possibilities for detecting unconventional New Physics signatures that have eluded detection so far. By leveraging advanced anomaly detection techniques and high-performance computing, both pioneered by the project's PI, this real-time system will capture and analyse events that are currently missed by the existing trigger system. This novel approach will enable CMS to explore the full landscape of collision data with unprecedented sensitivity. If successful, it could revolutionise how we search for New Physics and lead to novel discoveries that reshape our understanding of fundamental forces. This high-risk, high-reward project is a bold LEAP forward in the hunt for New Physics at the LHC, fully exploiting the capabilities of CMS and unlocking its full potential for discovery. This project’s technological innovations on real time Machine Learning inference, have the potential to impact fields such as satellite communications and large telescope arrays, broadening the scope of its significance.