Deep Generative Modeling  

深度生成建模  2022



This book discusses the latest advances in deep probabilistic generative models.And it does so in a very accessible way. What makes this book special is that, likethe child who is building a tower of bricks to understand the laws of physics, thestudent who uses this book can learn about deep probabilistic generative modelsby playing with code. And it really helps that the author has earned his spurs byhaving published extensively in this field. It is a great to tool to teach this topic inthe classroom.What will the future of our field bring? It seems obvious that progress towardsAGI will heavily rely on unsupervised learning. It’s interesting to see that thescientific community seems to be divided into two camps: the “scaling camp”believes that we achieve AGI by scaling our current technology to ever larger modelstrained with more data and more compute power. Intelligence will automaticallyemerge from this scaling. The other camp believes we need new theory and newideas to make further progress, such as the manipulation of discrete symbols (a.k.a.reasoning), causality, and the explicit incorporation of common-sense knowledge

这本书讨论了深层概率生成模型的最新进展。

它以一种非常容易理解的方式实现了这一点。这本书的特别之处在于

为了理解物理定律而建造砖塔的孩子

使用本书的学生可以了解深层概率生成模型

通过玩代码。这真的很有帮助，作者通过

在这个领域发表了大量的著作。这是一个很好的工具来教这个话题

教室。

我们这个领域的未来会带来什么？似乎很明显，朝着

AGI将严重依赖无监督学习。有趣的是看到

科学界似乎分为两个阵营：“规模化阵营”

相信我们通过将现有技术扩展到更大的型号来实现AGI

接受过更多数据和计算能力的训练。智能将自动

从这种规模中脱颖而出。另一个阵营认为我们需要新理论和新方法

取得进一步进展的想法，例如对离散符号的操作（又称。

推理），因果关系，以及常识知识的明确结合