The blackbox nature of statistical and generative AI has gained significant attention, but it requires greater explainability and safety, especially in critical fields like health, mental health and wellbeing, and crisis management. This course introduces techniques, systems, and measures for Neurosymbolic AI, leveraging knowledge in graphs and guidelines to enable machines to develop cognitive abilities necessary for these domains. It will guide you through the distinct stages of integrating knowledge into AI, emphasizing the differences between inferences from statistical AI and symbolic AI methodologies.
Highlighting significant advancements in statistical AI, including the popular Transformer models and attention mechanisms like BERT and GPT, this course offers a thorough and enlightening exploration of topics such as weakly supervised, distantly supervised, and unsupervised learning, along with their knowledge-enhanced counterparts. It delves into post-hoc and ante-hoc explainability and safety with knowledge, the integration of various forms of knowledge, and deeper levels of integration.
Technically, the course will explore cutting-edge areas such as reinforcement learning and policy gradients, zero-shot learning, active learning, and model fusion. Practically, it delves into innovations applied to real-world applications like conversational systems, mental health, and edge computing, featuring advancements by industry leaders like OpenAI and Google DeepMind.