2015/11/20 by Hristo S. Paskov, John C. Mitchell, Paskov, Hristo S. +3
Biochemistry, Genetics and Molecular Biology · Computer Science · #Algorithms and Data Compression #Embedded Systems Design Techniques #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning and Algorithms #Machine Learning in Bioinformatics #Numerical Methods and Algorithms
paper · pdf · doi:10.48550/arxiv.1511.06606
openalex publication_date 2015/11/20 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
We propose `Dracula', a new framework for unsupervised feature selection from\nsequential data such as text. Dracula learns a dictionary of n-grams that\nefficiently compresses a given corpus and recursively compresses its own\ndictionary; in effect, Dracula is a `deep' extension of Compressive Feature\nLearning. It requires solving a binary linear program that may be relaxed to a\nlinear program. Both problems exhibit considerable structure, their solution\npaths are well behaved, and we identify parameters which control the depth and\ndiversity of the dictionary. We also discuss how to derive features from the\ncompressed documents and show that while certain unregularized linear models\nare invariant to the structure of the compressed dictionary, this structure may\nbe used to regularize learning. Experiments are presented that demonstrate the\nefficacy of Dracula's features.\n