Deep Learning and Reasoning - 2018

Explore the intersection of deep learning and logical reasoning in this lecture by William Cohen from Google AI. Delve into TensorLog, a probabilistic first-order logic system that enables integration of symbolic reasoning with neural learning methods. Learn how TensorLog compiles inference to differentiable functions in neural network infrastructures like Tensorflow, allowing for parameter learning in probabilistic logic using high-performance deep learning frameworks. Discover applications of TensorLog in diverse tasks such as semi-supervised learning for network data, question-answering against knowledge bases, and relational learning. Gain insights from Cohen's extensive experience in machine learning, information integration, and statistical relational learning as he discusses the challenges and opportunities in combining logical reasoning with neural systems for advancing artificial intelligence.