An Online Learning Algorithm for Bilinear Models

Authors: Yuanbin Wu, Shiliang Sun

ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on two sequential labelling tasks give positive results. We conduct experiments on two sequential labelling tasks: word segmentation and text chunking.
Researcher Affiliation Academia Yuanbin Wu YBWU@CS.ECNU.EDU.CN Shiliang Sun SLSUN@CS.ECNU.EDU.CN Shanghai Key Laboratory of Multidimensional Information Processing Department of Computer Science and Technology, East China Normal University
Pseudocode Yes Algorithm 1 Blockwise Coordinate Descent. Algorithm 2 Online Learning of the Bilinear Model.
Open Source Code No The paper does not provide concrete access to source code for the methodology described.
Open Datasets Yes The data set is from the second SIGHAN Backoff (Emerson, 2005). We conduct experiments on the Co NLL Sharedtask 2000 (Sang & Buchholz, 2000).
Dataset Splits Yes The value R is selected on a dev set (10% of the pku training set).
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts) used for running its experiments.
Software Dependencies No The paper mentions software like 'SVM solver' and 'CRF' (crf2, crf1 from crfpp.googlecode.com/) but does not provide specific version numbers for these software components.
Experiment Setup Yes The algorithms for comparison are: bol which is the proposed method with T = 20, C = 1, R = 4, bcd which is the blockwise coordinate descent with SVM solver in Shalev-Shwartz & Singer (2006) (C = 1). We also compare state-of-the-art online linear model sp which is the structured perceptron with T = 20, learning rate C = 1.