The presence of multiple overlapping signals in the data stream of the Einstein Telescope and LISA severely decreases the efficacy of the classical matched filtering technique, making its computational cost unmanageable. The development of novel gravitational-wave data analysis methods during the next decade is imperative, and the use of machine-learning techniques is currently the most promising path. In the last few years, several studies have demonstrated that deep learning models generate posteriors of inferred source properties from current-generation detector observed data that are indistinguishable from those produced via standard methods. The primary goal of the “Gravitational-Wave Signal Searches” work package is to extend these methodologies to signals observed with future detectors, something not trivial due to the different signal morphology and frequency content. Another objective is the development of methods that increase the value of the Einstein Telescope, primarily, as part of a network of multi-messenger observatories, though the development of algorithms that allow for early, pre-merger alerts.