时间序列和自然语言处理的序列模型

课程概况

This course is an introduction to sequence models and their applications, including an overview of sequence model architectures and how to handle inputs of variable length.

• Predict future values of a time-series
• Classify free form text
• Address time-series and text problems with recurrent neural networks
• Choose between RNNs/LSTMs and simpler models
• Train and reuse word embeddings in text problems

You will get hands-on practice building and optimizing your own text classification and sequence models on a variety of public datasets in the labs we’ll work on together.

Prerequisites: Basic SQL, familiarity with Python and TensorFlow

课程大纲

Working with Sequences

In this module, you’ll learn what a sequence is, see how you can prepare sequence data for modeling, and be introduced to some classical approaches to sequence modeling and practice applying them.

Recurrent Neural Networks

In this module, we introduce recurrent neural nets, explain how they address the variable-length sequence problem, explain how our traditional optimization procedure applies to RNNs, and review the limits of what RNNs can and can’t represent.

Dealing with Longer Sequences

In this module we dive deeper into RNNs. We’ll talk about LSTMs, Deep RNNs, working with real world data, and more.

Text Classification

In this module we look at different ways of working with text and how to create your own text classification models.

Reusable Embeddings

Labeled data for our classification models is expensive and precious. Here we will address how we can reuse pre-trained embeddings to make our models with TensorFlow Hub.

Encoder-Decoder Models

In this module, we focus on a sequence-to-sequence model called the encoder-decoder network to solve tasks, such as Machine Translation, Text Summarization and Question Answering.

Summary

In this final module, we review what you have learned so far about sequence modeling for time-series and natural language data.