Efficient Sequence Regression by Learning Linear Models in All-Subsequence Space
|Title:||Efficient Sequence Regression by Learning Linear Models in All-Subsequence Space||Authors:||Gsponer, Severin
|Permanent link:||http://hdl.handle.net/10197/9054||Date:||22-Sep-2017||Abstract:||We present a new approach for learning a sequence regressionfunction, i.e., a mapping from sequential observations to a numericscore. Our learning algorithm employs coordinate gradient descent andGauss-Southwell optimization in the feature space of all subsequences.We give a tight upper bound for the coordinate wise gradients of squarederror loss that enables ecient Gauss-Southwell selection. The proposedbound is built by separating the positive and the negative gradients ofthe loss function and exploits the structure of the feature space. Extensiveexperiments on simulated as well as real-world sequence regressionbenchmarks show that the bound is eective and our proposed learningalgorithm is ecient and accurate. The resulting linear regression modelprovides the user with a list of the most predictive features selected duringthe learning stage, adding to the interpretability of the method.||Funding Details:||Science Foundation Ireland||Type of material:||Conference Publication||Keywords:||Machine learning; Statistics||Language:||en||Status of Item:||Peer reviewed||Conference Details:||The European Conference on Machine Learning & Principles and Practice of Knowledge Discovery in Databases, Skopje, Macedonia 18-22 September 2017|
|Appears in Collections:||Insight Research Collection|
Show full item record
This item is available under the Attribution-NonCommercial-NoDerivs 3.0 Ireland. No item may be reproduced for commercial purposes. For other possible restrictions on use please refer to the publisher's URL where this is made available, or to notes contained in the item itself. Other terms may apply.