A New View of Automatic Relevance Determination
By David Paul Wipf, Srikantan S. Nagarajan. 2007
405 citations recorded by the source corpus.
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
- Variational Inference: A Review for Statisticians (2016, neutral): “ance determination (ARD) assigns a separate prior for each regression coefficient; the idea is to automatically shrink irrelevant coefficients during inference (MacKay, 1992; Neal, 2012; Tipping, 2001; Wipf and Nagarajan, 2008). ARD works by pairing the prior precision of each regression coefficient wi”
- Optimization with Sparsity-Inducing Penalties (2011, methodology): “, maximum likelihood estimation of hyperparameters), iterative methods based on DC programming may be efficiently used [125].”
- Expectation propagation for neural networks with sparsity-promoting priors (2013, methodology): “In connection with linear models, various computationally more efficient algorithms have been proposed for determining marginal likelihood based point estimates for the relevance parameters (Tipping, 2001; Qi et al., 2004; Wipf and Nagarajan, 2008).”
- Linked independent component analysis for multimodal data fusion (2011, methodology): “To adapt to different scalings of the signal in eachmodality, an ARD prior (Wipf and Nagarajan, 2008) is used on the modality-courses (W).”
- Sparse Signal Recovery With Temporally Correlated Source Vectors Using Sparse Bayesian Learning (2011, methodology): “On the other hand, due to the works in [25], [27], [54], which connected SBL algorithms to traditional convex relax ation methods such as Lasso [37] and Basis Pursuit Denoising [38], it was found thatλ is functionally the same as the regularization parameters of those convex relaxation algorit hms.”
- A unified Bayesian framework for MEG/EEG source imaging (2009, unknown): “…(41) over S. Additionally, this duality can be extended the other way as well; γ-MAP can be expressed explicitly in S-space as a proper S-MAP method (Wipf and Nagarajan, 2008). solution to b = Ls satisfies jjsjj 0 Vjjs 0 jj 0 , meaning no solution with fewer number of equivalent dipoles could…”
- Probabilistic Multi-Task Feature Selection (2010, methodology): “It is noted that the automatic relevance determination (ARD) prior [9, 3, 26] comes as a special case under this interpretation.”
- Multitask Compressive Sensing (2009, background): “2While the sparsity analysis provided here is largely following that of RVM [19], which is intuitive and conceptual, a more recent and rigorous analysis of sparse Bayesian learning and its superior performance on sparse representation can be found at [37], [38].”
- Enhancing Sparsity by Reweighted ℓ1 Minimization (2007, methodology): “As a second example, Wipf and Nagarajan [42] have proposed an iterative reweighted 1 minimization algorithm for a feature selection problem known as automatic relevance determination (ARD).”
- EEGLAB, SIFT, NFT, BCILAB, and ERICA: New Tools for Advanced EEG Processing (2011, methodology): “…(3) Independent Modulators [38] (vii) Regularized and variational Bayesian logistic regression and sparse Bayesian logistic regression [39, 40] (vi) Baseline filtering (4) Multiband-CSP [41] (1) Hierarchical kernel learning [42] (vii) Resampling (5) Multi-Model Independent component…”
- Iterative Reweighted 1 and 2 Methods for Finding Sparse Solutions (2010, background): “While not discussed in [28], g SBL (x) is a general penalty function that only need have α = λ to obtain equivalence with SBL; other selections may lead to better performance.”
- Sparse Bayesian multi-task learning for predicting cognitive outcomes from neuroimaging measures in Alzheimer's disease (2012, methodology): “Our motivation of connecting T-MSBL-FP to existing algorithms is inspired by the work in [20, 25].”
- Image Super-Resolution Via Sparse Representation (2010, background): “Indeed, for face super-resolution, we can deal with lower resolution input images.”
- Smart Regularization of Deep Architectures (2015, methodology): “For example we could employ a Horsehoe distribution [48], an Automatic Relevance Determination [49, 50] framework, or a Bayesian Lasso [51] framework.”