Despite an exhaustive search program, there remains no conclusive evidence of physics beyond the Standard Model at the LHC. Machine learning (ML) is one of the innovative tools that could break new ground in this quest. This talk will explore three distinct approaches that can make physics analyses more precise, more sensitive, and more agnostic. First, we introduce DeepSF, a method that improves the shape agreement of data and simulation in the context of bottom-quark jet-tagging at the LHC. Next, we examine a reinforcement learning approach that increases the accuracy of matching partons to jets, thereby enabling sensitivity in multiple-jet final states. Lastly, we present HyLAnD, a new anomaly detection technique that employs a hybrid concept of unsupervised and semi-supervised learning to agnostically identify anomalous signals. Together, these innovative ML methods not only enhance our current capabilities but can also pave the way for future discoveries.