How does machine learning detect zero-day vulnerabilities

Question

How can machine learning analyze software behavior to detect vulnerabilities before they are known publicly, and how does it identify patterns and anomalies that indicate potential zero-day threats?