How can we verify identity using unstructured data from a user device? While biometrics like fingerprinting and facial recognition are often used for authentication, research around natural language processing has found people's use of language as uniquely identifying.

In this session, we will discuss multiple facets of language modeling:

• Efficacy on different kinds of unstructured text within a corporate network
• As a technique to detect anomalous user activity, compromised accounts, and stolen credentials
• As an integral part of a cybersecurity program in addition to UEBA and risk-adaptive protection

Every single security company is talking about how they are using machine learning—as a security company you have to claim artificial intelligence to be even part of the conversation. However, this approach can be dangerous when we blindly rely on algorithms to do the right thing. Rather than building systems with actual security knowledge, companies are using algorithms that nobody understands and, in turn, discovering wrong insights.

In this session, we will discuss:

• Limitations of machine learning and issues of explainability
• Where deep learning should never be applied
• Examples of how the blind application of algorithms can lead to wrong results