Explore a 17-minute conference talk from USENIX Enigma 2023 that delves into the challenges of applying machine learning to cybersecurity. Discover the complexities of dataset definition, class imbalance, adversarial machine learning, and concept drift in security applications. Learn about evaluation problems such as delayed labels and their impact on model training and testing. Gain insights into how existing solutions may fail under certain circumstances and explore potential fixes for these issues. Benefit from the speaker's goal of providing direction for future cybersecurity researchers and practitioners applying machine learning to their problems.