Active Manifold Learning via Gershgorin Circle Guided Sample Selection
Authors: Hongteng Xu, Hongyuan Zha, Ren-Cang Li, Mark Davenport
AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We test our active manifold learning algorithm on both regression and classification tasks. The active learning methods include: randomly labeling; (Random), two manifold related methods: the landmark method in (De Silva and Tenenbaum 2004) (Landmark) and the harmonic function method3 in (Zhu, Lafferty, and Ghahramani 2003) (Harmonic); and our Gershgorin circle guided method (GC). The semi-supervised manifold learning algorithms for regression include the least squares method in (Yang et al. 2006) (LS) and the spectral method in (Zhang, Zha, and Zhang 2008) (Spectral). Like (Yang et al. 2006; Zhang, Zha, and Zhang 2008; Zhang, Wang, and Zha 2012), we compute the alignment matrix by LTSA. For classification, graph regularized sparse coding model (Graph SC, Eq. (4)) is applied. After learning sparse codes of samples, we label some samples by various active learning methods and train SVM classifier by labeled sparse codes. |
| Researcher Affiliation | Academia | 1School of ECE, Georgia Institute of Technology, Atlanta, GA, USA 2College of Computing, Georgia Institute of Technology, Atlanta, GA, USA 3Software Engineering Institute, East China Normal University, Shanghai, China 4Department of Mathematics, University of Texas at Arlington, Arlington, TX, USA |
| Pseudocode | Yes | Circle Deletion Algorithm (CDA) 1. For N N matrix M, compute its Gershgorin circles {Ci}N i=1. [...] Active Manifold Learning Input: Sample set X RD N; The number of labeled samples we can select, L; Algorithm: 1. Given the K-NN graph of X RD N, compute the alignment matrix Φ by a certain method. [...] |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. It does not contain links to code repositories or explicit statements about code availability. |
| Open Datasets | Yes | The first data set we use is the face data from (Tenenbaum, De Silva, and Langford 2000). Another data set we used is from (Rahimi, Darrell, and Recht 2005), which shows a subject moving his arms. [...] The ARface data set (Martinez 1998) contains over 4000 frontal view faces corresponding to 126 people s faces. [...] The Extended Yale B data set contains 2414 frontal face images of 38 persons. [...] The Caltech101 data set (Fei-Fei, Fergus, and Perona 2007) contains 9144 images from 101 object classes and a background class. |
| Dataset Splits | No | The paper does not provide specific dataset split information for validation, such as exact percentages, sample counts, or explicit mention of a validation set or cross-validation setup. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details, such as library or solver names with version numbers, needed to replicate the experiments. It mentions "SVM classifier" but without a version. |
| Experiment Setup | Yes | In each trial, we randomly select 600 samples and construct a K-NN graph, K = 8. The parameter of the spectral method is set to λ = N L. ... Repeating the test 100 times in both noise-free and noisy cases (Gaussian noise with zero mean and variance σ2 = 0.01)... |