You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Demystifying Deep Convolutional Neural Networks 文章推导了多隐藏层的公式,图文并茂的讲解了原理,并且用Matlab编码实现,强烈推荐。
w2v Principle and Source Code Analysis
word2vec is a short 700 lines of code, but the depth of the programing skill is breathtaking. The author, Tomas Mikolov, is supposed to be an algorithm contest player, judging by his code style. Programming skills that can be learned from this code are: Handling of input parameters, Accelerate sigmoid function calculations with a segmented lookup table, Weighted sampling, Tips for reading files, Dynamically control the size of the hash table, Building Huffman trees, Splitting files by thread, Random number generation methods, Randomizing window size, Adaptive learning rate.
"Mathematical principles in word2vec" This article explains in detail the machine learning principles used in this code: Logistic regression, language models, optimization goals, formula derivation, gradient update, a variety of detailed techniques are available, formula derivation and coding consistency is very high. Understand w2v, read this article is enough.