Datum-Wise Classification: A Sequential Approach to Sparsity
We propose a novel classification technique whose aim is to select an appropriate representation for each datapoint, in contrast to the usual approach of selecting a representation encompassing the whole dataset. This datum-wise representation is found by using a sparsity inducing empirical risk, which is a relaxation of the standard L 0 regularized risk. The classification problem is modeled as a sequential decision process that sequentially chooses, for each datapoint, which features to use before classifying. Datum-Wise Classification extends naturally to multi-class tasks, and we describe a specific case where our inference has equivalent complexity to a traditional linear classifier, while still using a variable number of features. We compare our classifier to classical L 1 regularized linear models (L 1-SVM and LARS) on a set of common binary and multi-class datasets and show that for an equal average number of features used we can get improved performance using our method.
KeywordsFeature Selection Sequential Approach Reward Function Empirical Risk Policy Iteration
Unable to display preview. Download preview PDF.
- 1.Tibshirani, R.: Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society. Series B (January 1994)Google Scholar
- 2.Sutton, R., Barto, A.: Reinforcement Learning. MIT Press, Cambridge (1998)Google Scholar
- 5.Har-Peled, S., Roth, D., Zimak, D.: Constraint classification: A new approach to multiclass classification. Algorithmic Learning Theory, 1–11 (2002)Google Scholar
- 6.Lagoudakis, M.G., Parr, R.: Reinforcement learning as classification: Leveraging modern classifiers. In: ICML 2003 (2003)Google Scholar
- 9.Girgin, S., Preux, P.: Feature discovery in reinforcement learning using genetic programming. In: O’Neill, M., Vanneschi, L., Gustafson, S., Esparcia Alcázar, A.I., De Falco, I., Della Cioppa, A., Tarantino, E. (eds.) EuroGP 2008. LNCS, vol. 4971, pp. 218–229. Springer, Heidelberg (2008)CrossRefGoogle Scholar
- 10.Gaudel, R., Sebag, M.: Feature Selection as a One-Player Game. In: ICML (2010)Google Scholar
- 12.Ertin, E.: Reinforcement learning and design of nonparametric sequential decision networks. In: Proceedings of SPIE, pp. 40–47 (2002)Google Scholar
- 14.Póczos, B., Abbasi-Yadkori, Y., Szepesvári, C., Greiner, R., Sturtevant, N.: Learning when to stop thinking and do something! In: ICML 2009, pp. 1–8 (2009)Google Scholar
- 15.Dulac-Arnold, G., Denoyer, L., Gallinari, P.: Text Classification: A Sequential Reading Approach. In: ECIR, pp. 411–423 (2011)Google Scholar
- 16.Preda, M.: Adaptive building of decision trees by reinforcement learning. In: Proceedings of the 7th WSEAS, pp. 34–39 (2007)Google Scholar