Outlier detection by active learning

By Naoki Abe, Bianca Zadrozny, J. Langford. Knowledge Discovery and Data Mining, 2006

368 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/71df4b737c35fb3d05f44036419e78b5330b580f

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

Citations in the corpus

Contextual Outlier Interpretation (2017, result): “learning has also been implied in some previous work [1, 42].”
A Unifying Review of Deep and Shallow Anomaly Detection (2020, background): “To improve over artificial uniform sampling, more informed sampling strategies have been proposed [341], such as resampling schemes [342], manifold sampling [343], and sampling based on local density estimation [344], [345], as well as active learning strategies [346]–[348].”
Learning to Locate Relative Outliers (2011, methodology): “The artificial outliers here are created by using a uniform distribution U that is defined within a bounded subspace whose minimum and maximum are limited to be 10% beyond the observed minimum and maximum from the training data (Abe et al., 2006).”
Extreme Value Theory for Open Set Classification - GPD and GEV Classifiers (2018, methodology): “Machine learning techniques have been applied successfully to both tasks, improving considerably their performance [12]–[14].”
iBAT: detecting anomalous taxi trajectories from GPS traces (2011, methodology): “The proposed methods range from supervised approaches [1] to distance-based [12, 3], density-based [5], and model-based methods [11], each having a specific formulation of the problem with different anomaly measures and notions.”
Anomaly detection: A survey (2009, background): “A number of techniques have been proposed that inject artificial anomalies in a normal data set to obtain a labeled training data set [Theiler and Cai 2003; Abe et al. 2006; Steinwart et al. 2005].”
A Comparative Evaluation of Unsupervised Anomaly Detection Algorithms for Multivariate Data (2016, methodology): “To reduce the number of anomalies, we select the class 1 as normal and apply a stratified sampling for the classes 2, 3, 5, 6 and 7, similar to [67, 68].”
Isolation-Based Anomaly Detection (2012, methodology): “Notable examples are: classification-based methods [Abe et al. 2006], Replicator Neural Network (RNN) [Williams et al.”
An instance level analysis of data complexity (2014, background): “Keywords Instance hardness · Dataset hardness · Data complexity”
Outlier Detection (2008, background): “Tang et al. [55] defined the connectivity-based outlier factor that compares favorably with LOF at low density regions.”
Graph based anomaly detection and description: a survey (2014, neutral): “…[Saltenis, 2004], clustering-based [He et al., 2003, Lieto et al., 2008, Miller and Browning, 2003, Wang et al., 2012c], classification-based [Abe et al., 2006, Hempstalk et al., 2008, Janssens et al., 2009], information theory-based [Ando, 2007, Böhm et al., 2009, Smets and Vreeken, 2011],…”
Outlier Analysis (2013, methodology): “A selective sampling method which uses active learning in the context of outlier detection is proposed in [1].”
Interpreting and Unifying Outlier Scores (2011, background): “Omitting the threshold p and reporting the fraction p x of data objects o ∈ D where dist(x, o) > D results in an outlier score for x in the range [0] [1].”
Hierarchical Density Estimates for Data Clustering, Visualization, and Outlier Detection (2015, methodology): “…repository [Bache and Lichman 2013], with adaptations for the evaluation of outlier detection methods as it is common practice in the literature [Abe et al. 2006; Yang et al. 2008; Kriegel et al. 2008; Zhang et al. 2009; Keller et al. 2012; Zimek et al. 2013a]: In the Lymphography dataset, the…”
Ensembles for unsupervised outlier detection: challenges and research questions a position paper (2014, methodology): “This procedure is sometimes called “down sampling” and has been used, with different variants, in many studies designing new outlier detection methods [1; 73; 42; 74; 36; 15; 14].”
Isolation Forest (2008, methodology): “Apart from the key difference of isolation versus pro-ﬁling, iForest is distinguished from existing model-based [11, 1, 5], distance-based [6] and density-based methods [4] in the follow ways: • The isolation characteristic of iTrees enables them to build partial models and exploit sub-sampling to…”