Скачать или смотреть The Story of Neural Networks

The provided video offers an overview of the development and significance of various neural network architectures within the field of deep learning. Specifically, it focuses on AlexNet, an influential 2012 deep convolutional neural network (CNN) that achieved breakthrough performance in image classification. Also explores the broader history of Recurrent Neural Networks (RNNs), detailing their origins in neuroscience and physics, various configurations like Bidirectional RNNs, and training methods such as Backpropagation Through Time (BPTT). A specialized RNN architecture, the Long Short-Term Memory (LSTM) network, is discussed in detail as a solution to the vanishing gradient problem in traditional RNNs. Finally, modern advancements like the Transformer model, which dispenses with recurrence and convolutions using only attention mechanisms, and Residual Networks (ResNets), which use identity mapping for training ultra-deep networks, are highlighted as key innovations in sequence transduction and computer vision.