Decision tree-based anomaly detection on FPGA

Anomaly detection using a decision trees on field programmable gate array, FPGA, is discussed. The decision tree-based autoencoder is trained using importance sampling of the vector space of input variables. Using LHC-inspired simulated data samples inference is made at 30 ns using minimal resource utilization, which is competitive to the neural network-based models, such as variational autoencoders, that are simplified to implement on FPGA-based real-time trigger systems.