Attacking Deep Learning-Based NLP Systems

Explore the security vulnerabilities of Deep Learning-based Natural Language Processing (NLP) systems in this insightful conference talk from BSidesSF 2019. Delve into the world of Word Embeddings and their crucial role in modern NLP systems. Discover how the process of data gathering, training word embeddings, and integrating them into NLP systems can be exploited from a security perspective. Learn about practical demonstrations of influencing these systems through training data manipulation and methods of injecting them into real-world applications. The talk covers key topics including the importance of NLP, Word Embeddings, Sentiment Analysis, and Medical Chatbots, while also discussing potential defenses against such attacks. Gain valuable insights into this often overlooked aspect of NLP security and participate in a Q&A session to further your understanding.