Abstract
In this paper, we present a supervised learning technique that utilizes artificial neural networks to design new collective entanglement witnesses for two-qubit and qubit-qutrit systems. Machine-designed collective entanglement witnesses allow for continuous tuning of their sensitivity and selectivity. These witnesses are, thus, a conceptually novel instrument allowing to study the sensitivity vs. selectivity trade-off in entanglement detection. The chosen approach is also favored due to its high generality, lower number of required measurements compared to quantum tomography, and potential for superior performance with regards to other types of entanglement witnesses. Our findings could pave the way for the development of more efficient and accurate entanglement detection methods in complex quantum systems, especially considering realistic experimental imperfections.