Toward Optimal Feature Selection

By Daphne Koller, Mehran Sahami. 1996

How researchers see this paper

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

Citations in the corpus

Fast Conditional Independence Test for Vector Variables with Large Sample Sizes (2018, neutral): “Probabilistic dependence – conditional or not – is often taken as an indicator of informational exchange (see e.g. Cover and Thomas [2012], Chapter 2), causal connection [Dawid, 1979, Spirtes et al., 2000] or simply as tool for prediction and diagnosis [Koller and Sahami, 1996].”
Are Key-Foreign Key Joins Safe to Avoid when Learning High-Capacity Classifiers? (2017, background): “The trade-off between feature redundancy and relevancy is well-studied [16, 50, 24].”
Feature selection algorithms: a survey and experimental evaluation (2002, background): “The cost is , although in practice it demands more computation than forward [ 12 ].”
A hybrid genetic algorithm for feature selection wrapper based on mutual information (2007, background): “Feature selection involves ﬁnding a subset of features to improve prediction accuracy or decrease the size of the structure without signiﬁcantly decreasing the prediction accuracy of the classiﬁer built by using only the selected features (Ahmad and Dey, 2005; Koller and Sahami, 1996).”
Feature Selection for Classification: A Review (2014, methodology): “√ n nj − √ nj n if xi ∈ cj , − √ nj n otherwise (0.4) Mutual Information based on Methods [37, 74, 52]: Due to its computational efficiency and simple interpretation, information gain is one of the most popular feature selection methods.”
Feature selection via regularized trees (2012, background): “Theoretical ly, the optimal feature subset should be a minimal feature set without loss of predictive information and can be formulate d as a Markov blanket of Y (MB(Y )) [21], [22].”
Supervised feature selection via dependence estimation (2007, background): “Examples include the leave-one-out error bound of SVM (Weston et al., 2000) and the mutual information (Koller & Sahami, 1996).”
Markov blanket-embedded genetic algorithm for gene selection (2007, methodology): “In 1996, Koller and Sahami [37] proposed a cross-entropy based technique, known as Markov Blanket for identifying redundant and irrelevant features.”
On Azadkia–Chatterjee’s conditional dependence coefficient (2021, background): “Testing conditional independence is nowadays a routine task in graphical modeling (Maathuis et al., 2019), causal discovery (Peters et al., 2017), feature selection (Koller and Sahami, 1996), and many other statistical applications.”
Autonomous Agents and Multi-agent Systems Manuscript No. Learning Potential Functions and Their Representations for Multi-task Reinforcement Learning (unknown): “FS-TEK’smainloopisbasedonaniterative backwardelimination procedure(e.g. [28,32]).”
Incremental wrapper-based gene selection from microarray data for cancer classification (2006, unknown): “In [21] it is stated that the cardinality of set M must be small and fixed.”
Many-Layered Learning (2002, methodology): “Feature selection algorithms typically fall into two categ ories: wrapper methodssuch as those described by John, Kohavi and Pfleger (1994) or Caruana and Freitag (1994) and filter methods such as those described by Hall and Smith (1996) or Koller and Sahami (1996) .”
Causal Feature Selection (2007, background): “[20, 7] contained many promising ideas and the latter was successfully applied in real data, however they were not guaranteed to find the actual Markov blanket and nor could they be scaled to thousands of variables.”
A review of feature selection techniques in bioinformatics (2007, methodology): “Recently, the avenue of FS techniques for coding potential prediction was further pursued by (Saeys et al ., 2007), who combined different measures of coding potential prediction, and then used the Markov blanket multivariate filter approach (MBF) to retain only the relevant ones.”
HITON: A Novel Markov Blanket Algorithm for Optimal Variable Selection (2003, methodology): “In the reported experiments we did not include any of the previous methods for inducing MB(T) (most notably the highly-cited Koller-Sahami algorithm [11], but also the ones in [1, 2, 10]) because they are computationally intractable for datasets as large as the ones used in our evaluation.”
Time and sample efficient discovery of Markov blankets and direct causal relations (2003, methodology): “MMMB is compared with all previously known algorithms for inducing Markov Blankets, namely the Incremental Association Markov Blanket (IAMB) algorithm [11], the Grow-Shrink (GS) algorithm [5], and the Koller-Sahami algorithm (KS) [4].”
Consistency-based search in feature selection (2003, methodology): “Information measure is employed in [2] [8] [23] [40].”