Feature Generation Using General Constructor Functions

By Shaul Markovitch, Dan Rosenstein. 2002

176 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computational Theory and Mathematics, Information Systems

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computational Theory and Mathematics

In Information Systems

Citations in the corpus

Benchmark and Survey of Automated Machine Learning Frameworks (2019, background): “with different heuristics to explore the feature space in a best-first way (Markovitch and Rosenstein, 2002).”
Survey on Automated Machine Learning (2019, methodology): “y calculate the performance of the new feature set: Early approaches combined beam search in combination 13 Zoller and Huber with dierent heuristics to explore the feature space in a best-rst way (Markovitch and Rosenstein, 2002). More recently, greedy search (Dor and Reich, 2012; Khurana et al., 20”
Computational Phenotype Discovery Using Unsupervised Feature Learning over Noisy, Sparse, and Irregular Clinical Data (2013, background): “Related work that reaches back fifty years [15–18] investigates supervised feature learning, where the expert selects the learning task, a set of raw input variables and the outcome labels, and then the algorithm tries to construct features from the input variables either individually [12,15,16] or…”
Supervised Machine Learning: A Review of Classification Techniques (2007, methodology): “Classification accuracy of rule learning algorithms can be improved by combining features (such as in decision trees) using the background knowledge of the user (Flach and Lavrac, 2000) or automatic feature construction algorithms (Markovitch and Rosenstein, 2002).”
Recursive Feature Generation for Knowledge-based Learning (2018, methodology): “The FICUS algorithm (Markovitch & Rosenstein, 2002) allows the use of any combinational feature generation technique, based on a given set of constructor functions.”
Feature Engineering for Predictive Modeling using Reinforcement Learning (2017, methodology): “Note that FICUS is more general than a number of less recent approaches (Ragavan et al. 1993; Bagallo and Haussler 1990; Yang, Rendell, and Blix 1991; Matheus and Rendell 1989; Hu and Kibler 1996).”
Data Preprocessing for Supervised Leaning (2007, unknown): “While FICUS [36] is similar in some aspects to some of the existing feature construction algorithms (such as GALA), its main strength and contribution are its generality and flexibility.”
Machine learning: a review of classification and combining techniques (2006, background): “Classification accuracy of rule learning algorithms can be improved by combining features (such as in decision trees) using the background knowledge of the user (Flach and Lavrac 2000) or automatic feature construction algorithms (Markovitch and Rosenstein 2002).”
Overcoming the Brittleness Bottleneck using Wikipedia: Enhancing Text Categorization with Encyclopedic Knowledge (2006, background): “Feature generation techniques, which search for new features that describe the target concept better than the ones supplied with the training instances, were found useful in a variety of machine learning tasks (Markovitch & Rosenstein 2002; Fawcett 1993; Matheus 1991).”
AutoLearn — Automated Feature Generation and Selection (2017, background): “The framework proposed in [12] improves over [9], but requires domain proficiency to select feature constructors.”
Distributional Word Clusters vs. Words for Text Categorization (2003, background): “Table 1 summarizes the results that were discussed in this section.”
FeatureInsight: Visual support for error-driven feature ideation in text classification (2015, background): “On problems where data is plentiful, features can sometimes be automatically generated [8], [9], [10] or selected from a candidate set [2], [4], [11], [12].”
Strengthening learning algorithms by feature discovery (2012, background): “They create new features by a variety of mechanisms including using high-order polynomial functions [24], non-linear combination of features [9,16] or using arithmetic and logic constructors adapted for each target dataset differently [13].”
Learning Feature Engineering for Classification (2017, methodology): “To perform automated feature engineering, some existing approaches adopt guided-search in feature space using heuristic feature quality measures (such as information gain) and other surrogate measures of performance [Markovitch and Rosenstein, 2002; Fan et al. , 2010].”
Feature Generation for Text Categorization Using World Knowledge (2005, methodology): “Feature generation (FG) techniques were found useful in a variety of machine learning tasks [Markovitch and Rosenstein, 2002; Fawcett, 1993; Matheus, 1991 ].”