tailieunhanh - Báo cáo khoa học: "Semi-supervised Learning for Natural Language Processing"
The amount of unlabeled linguistic data available to us is much larger and growing much faster than the amount of labeled data. Semi-supervised learning algorithms combine unlabeled data with a small labeled training set to train better models. This tutorial emphasizes practical applications of semisupervised learning; we treat semi-supervised learning methods as tools for building effective models from limited training data. An attendee will leave our tutorial with 1. | Semi-supervised Learning for Natural Language Processing John Blitzer Natural Language Computing Group Microsoft Research Asia Beijing China blitzer@ Xiaojin Jerry Zhu Department of Computer Science University of Wisconsin Madison Madison WI USA jerryzhu@ 1 Introduction The amount of unlabeled linguistic data available to us is much larger and growing much faster than the amount of labeled data. Semi-supervised learning algorithms combine unlabeled data with a small labeled training set to train better models. This tutorial emphasizes practical applications of semisupervised learning we treat semi-supervised learning methods as tools for building effective models from limited training data. An attendee will leave our tutorial with 1. A basic knowledge of the most common classes of semi-supervised learning algorithms and where they have been used in NLP before. 2. The ability to decide which class will be useful in her research. 3. Suggestions against potential pitfalls in semisupervised learning. 2 Content Overview Self-training methods Self-training methods use the labeled data to train an initial model and then use that model to label the unlabeled data and retrain a new model. We will examine in detail the cotraining method of Blum and Mitchell 2 including the assumptions it makes and two applications of co-training to NLP data. Another popular selftraining method treats the labels of the unlabeled data as hidden and estimates a single model from labeled and unlabeled data. We explore new methods in this framework that make use of declarative linguistic side information to constrain the solutions found using unlabeled data 3 . Graph regularization methods Graph regularization methods build models based on a graph on instances where edges in the graph indicate similarity. The regularization constraint is one of smoothness along this graph. We wish to find models that perform well on the training data but we also regularize so that .
đang nạp các trang xem trước