Non-Conceptual Knowledge Systems

This lecture explores the impact of deep learning on Digital Humanities, focusing on non-conceptual knowledge systems. It delves into the characteristics of articulated reasoning and the challenges of grounding real-world data into logical symbols. The presentation discusses the recent success of non-conceptual knowledge systems, such as AlexNet and Neural Machine Translation, and the limitations of recurrent neural networks. It also covers the advancements in transformers and the attention mechanism, highlighting the potential of BERT and GPT-3 models. The lecture concludes with a discussion on generative adversarial networks and the issues of vanishing gradients and mode collapse.