 Artificial intelligence in the security market is valued $4 billion and expected to grow to more than $34 billion by 2025. In these talks, I'll explain the reason for these growths and how artificial intelligence may be utilized for cyber security. While the rapid growths in the number of cyber attacks, the increasing sophistication and mounting impact along with data explosion has transformed cyber security into a big data problem that cannot be handed anymore by traditional security systems or by cyber experts. Well, traditional security systems match incoming intrusions with non-malicious behavior using predefined rules. Such technology cannot cope with unknown attack or with well hidden known threats. Artificial intelligence and specifically sophisticated machine learning algorithms can process big data efficiently to automatically detect and prevent known attack and predict unknown threat that's minimizing the need for security specialists. Supervised machine learning algorithms train on many examples of benign and malicious event learning their hidden patterns. Those patterning may then be used to accurately classify new events as malicious or benign learning to detect known attacks. Early detection algorithms are new types of other types of algorithms that train only on benign examples, thus learning how to detect new attacks that deviate from the learned benign patterns. Machine learning algorithms may be used for intrusion detections, assuming that each program leaves some footprints in the cyberspace. Those footprints form patterns that can then differentiate between benign and malicious programs. In a study of our labs, we learned the profile of mobile applications by their network traffic. We're then able to identify malicious versions of applications by looking at applications that whose behavior deviate from the learned profile. Machine learning can be also used for user authentication by learning legitimate behavior of users from the interaction, such as keystroke, touch gestures, movements, etc. An illich to meet users would deviate from the learned benign profile of the users. In a study in our labs, we learned the profiles of smartphone owners by continuously detecting, tracing their sensors from the phone. We're then able to detect a theft of a device in less than six seconds by comparing the behavior of the users with the learned profile. Machine learning can be used for attack attribution, automatically associating features of the attack with specific threat actors. The features of the attack that can be analyzed may be the target of the attack, the user's use, all kind of temporal and geographical features of the attack. The set of features to use and how to manipulate them is termed feature engineering and is typically performed by a user expert. Feature engineering is a major challenge for machine learning algorithms and is becoming a bottleneck. This is because cyber experts and data scientists need to estimate the set of features that would best represent the desired pattern. This is really a major challenge. Deep learning algorithms, which are a new emerging powerful technology that is a branch of machine learning that utilizes artificial neural network, minimizes the need for feature engineering as a self-learning ability of neural network can learn the interaction between the neurons and form the features. To sum up, machine learning and specifically deep learning algorithms are very beneficial and fundamental for cyber defense. Despite the many challenges, security is slowly maturing to become a standard for cyber security. Thank you for your attention.