This tutorial provides an overview of neural network models for natural language processing (NLP). It covers various neural network architectures, including feed-forward networks, convolutional networks, recurrent networks, and recursive networks. The tutorial discusses input encoding for NLP tasks, the computation graph abstraction for automatic gradient computation, and the application of neural networks in NLP. It also addresses the differences between sparse and dense feature representations, the use of embeddings, and the importance of non-linearities in neural networks. The tutorial emphasizes the benefits of dense vector representations over one-hot representations, and discusses the use of different non-linear activation functions such as sigmoid, tanh, hard tanh, and ReLU. It also covers output transformations such as softmax, and discusses loss functions used in training neural networks, including hinge loss, log loss, categorical cross-entropy loss, and ranking losses. The tutorial aims to provide NLP practitioners with the basic background, jargon, tools, and methodology needed to understand and apply neural network models to their work. It also highlights the importance of choosing appropriate architectures and parameters for different NLP tasks.This tutorial provides an overview of neural network models for natural language processing (NLP). It covers various neural network architectures, including feed-forward networks, convolutional networks, recurrent networks, and recursive networks. The tutorial discusses input encoding for NLP tasks, the computation graph abstraction for automatic gradient computation, and the application of neural networks in NLP. It also addresses the differences between sparse and dense feature representations, the use of embeddings, and the importance of non-linearities in neural networks. The tutorial emphasizes the benefits of dense vector representations over one-hot representations, and discusses the use of different non-linear activation functions such as sigmoid, tanh, hard tanh, and ReLU. It also covers output transformations such as softmax, and discusses loss functions used in training neural networks, including hinge loss, log loss, categorical cross-entropy loss, and ranking losses. The tutorial aims to provide NLP practitioners with the basic background, jargon, tools, and methodology needed to understand and apply neural network models to their work. It also highlights the importance of choosing appropriate architectures and parameters for different NLP tasks.