The Research Of IB Algorithm Based On Selective Model

IB(Information Bottleneck) method originated in the rate-distortion theory is an effective solution to complex optimization problems in data compression and information preservation. In the field of machine learning, IB method is used for pattern extraction, and many researchers proposed multiple versions of IB algorithm. When IB is used for Pattern extraction, IB algorithm analyses all data objects. However, in practical applications, the analysis of low-relevant or irrelevant data is not conducive to the pattern extraction, resulting in low precision of the data patterns, or unsuccessful excavation of hidden data pattern.To solve the above problems, this paper proposes a selective IB model-SLIB, corresponding objective function and the corresponding algorithm. There are many data objects of uncertain pattern structure in datasets. analyzing these data objects does not only result in the loss of large amounts of information, but also makes data pattern deviate hidden data pattern. Therefore, the SLIB model only analyses some data objects with a clear pattern structure, and filters uncertain others. Unlike the conventional IB objective function, the SLIB objective function is not only can achieve the selective analysis of the data, but also applies to co-occurrence data and non co-occurrence data. Optimizing the objective function by using a sequential "extraction-merger" strategy, the SLIB algorithm can quickly converge to a local optimal result. In artificial, co-occurrence and Euclidean space datasets, the experiment result show the SLIB algorithm ensures high recall, while the data pattern is higher precision than the original IB algorithms, and the original IB algorithms cannot min the same one.IB method has been successfully applied in many fields, and the SLIB algorithm in this paper can be applied to information retrieval, because it easily search the high-precision result. Besides, the SLIB algorithm cannot only be used for a wider variety of datasets, but also the SLIB model provides a new research clue for.IB method.