行业动态
www.freecodecamp.org ·
@Beau Carnes
动手复现深度学习演进：Geoffrey Hinton 的 AI 遗产课程
freeCodeCamp 发布了一门 27 小时免费课程，由 Mohammed Abrah 讲解，按历史脉络复现 Geoffrey Hinton 的关键工作，涵盖 Boltzmann Machines、反向传播、Deep Belief Networks、t-SNE、AlexNet、Dropout、知识蒸馏、Capsule Networks、Layer Normalization 和 Forward-Forward Algorithm，并用 PyTorch 提供教学实现与可视化实验。
AI推荐理由 freeCodeCamp 发布了一门 27 小时的免费课程，由 Mohammed Abrah 讲解，按时间线复现 Hinton 的关键论文，从 Boltzmann Machines、反向传播、Deep Belief Networks、t-SNE、AlexNet、Dropout、知识蒸馏、Capsule Networks、Layer Normalization 到 Forward-Forward Algorithm，并配有 PyTorch 教学实现。信息密度较高，属于可验证的公开课程资源，对学习深度学习历史与实现有实际用途；但它是课程推广性质的内容，并非新研究或新模型发布，也没有新的技术结论，因此兴趣分中等。
62
正文
Geoffrey Hinton is known as the "the Godfather of AI". We just published a course on the freeCodeCamp.org YouTube channel that will help you master modern neural networks by recreating the groundbreaking discoveries of Geoffrey Hinton. The course covers his key papers and you will follow the evolution of the ideas across decades, from Boltzmann Machines and backpropagation to Deep Belief Networks, t-SNE, AlexNet, Dropout, Knowledge Distillation, Capsule Networks, Layer Normalization, and the Forward-Forward Algorithm. Throughout this 27-hour course, Mohammed Abrah will explain the motivation behind each work, the problem it addressed, the core idea introduced by the paper, and then connect the theory to an educational implementation in PyTorch. The tutorial explores: The historical evolution of Geoffrey Hinton's research across deep learning How Boltzmann Machines introduced learning through energy-based models Learning representations through backpropagation Distributed representations and Adaptive Mixtures of Local Experts The Helmholtz Machine and the Wake-Sleep Algorithm Stochastic Neighbor Embedding and the foundations of t-SNE Deep Belief Networks and greedy layer-wise pretraining Deep Boltzmann Machines How Rectified Linear Units helped improve Restricted Boltzmann Machines ImageNet Classification with Deep Convolutional Neural Networks (AlexNet) Dropout as a method for reducing overfitting Knowledge Distillation and transferring knowledge between neural networks Layer Normalization Dynamic Routing Between Capsules Contrastive Learning of Visual Representations The Forward-Forward Algorithm Practical PyTorch implementations of the core ideas Visualizations and experiments that make the underlying algorithms easier to understand The connection between these historical ideas and modern AI architectures Watch the full course on the freeCodeCamp.org YouTube channel (27-hour watch).