Uncovering shared structures in multiclass classification

By Yonatan Amit, Michael Fink, Nati Srebro, Shimon Ullman. 2007

356 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

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 Computer Vision and Pattern Recognition

Citations in the corpus

Semi-Cyclic Stochastic Gradient Descent (2019, background): “…structure (Ando & Zhang, 2005), shared deep features (e.g. Caruana, 1997), shared kernel or low nuclearnorm structure (e.g. Argyriou et al., 2008; Amit et al., 2007), low-norm additive structure (e.g. Evgeniou & Pontil, 2004), or graph-based relatedness (e.g. Evgeniou et al., 2005; Maurer, 2006).”
A Singular Value Thresholding Algorithm for Matrix Completion (2008, methodology): “applied science such as machine learning [1, 3, 4], control [45] and computer vision, see [52].”
Optimization with Sparsity-Inducing Penalties (2011, background): “Using the trace norm as a convex surrogate for rank has many applications in control theory [48], matrix completion [1, 130], multi-task learning [109], or multi-label classification [4], where low-rank priors are adapted.”
Label-Embedding for Image Classification (2015, neutral): “Consequently, the main applications have been multiclass classiﬁcation with many classes[3], [66], [67], [4] and zero-shot learning [31], [43].”
Excess risk bounds for multitask learning with trace norm regularization (2012, methodology): “In this paper we consider multi-task learning with trace-norm regularization (TNML), a technique for which eﬃcient algorithms exist and which has been successfully applied many times (see e.g. [2, 4, 14, 15]).”
Implicit Regularization in Matrix Factorization (2017, background): “In particular, this setting covers problems including matrix completion (where Ai are indicators, [5]), matrix reconstruction from linear measurements [15] and multi-task training (where each column of X is a predictor for a different task and Ai have a single non-zero column, [2, 1]).”
Regularization Techniques for Learning with Matrices (2009, background): “Using WS(1) in such a case was previously suggested by Amit et al. (2007), who observed that empirically, the class WS(1) performs better than W2,2 when there is a shared structure between classes. They only gave empirical results and our analysis given in Corollary 23 provides a first rigorous regr”
On Better Exploring and Exploiting Task Relationships in Multitask Learning: Joint Model and Feature Learning (2018, background): “Multi-way classification is treated as a multi-task learning problem, where each task is a classification task of one digit against all the other digits [31, 40].”
Tensor Completion for Estimating Missing Values in Visual Data (2009, background): “Typically, µ is in the range [1 , 10] , if the data is normalized as above.”
Evaluation of deep convolutional nets for document image classification and retrieval (2015, background): “The goal of transfer learning is to facilitate learning on problems with insufficient training data, by taking advantage of shared structure in related problems [1].”
Sharing features between objects and their attributes (2011, background): “The focus has been on inter-class transfer between objects [10, 5, 22, 20] , and jointly training detectors so that features are shared across objects [23, 3, 29], which lends to better efficiency during detection.”
The Power of Convex Relaxation: Near-Optimal Matrix Completion (2009, methodology): “Index Terms— Duality in optimization, free probability, low-rank matrices, matrix completion, nuclear norm minimization, random matrices and techniques from random matrix theory, semideﬁnite programming.”
Large Scale Visual Recognition (2012, background): “Our problem is related to cost-sensitive classification and hierarchical classification [126, 127, 52, 53, 128, 26, 27, 29, 24, 31, 23, 28, 25, 22, 129, 130, 131].”
A convex formulation for learning shared structures from multiple tasks (2009, methodology): “Recently, there is growing interest in studying multi-task learning in the context of feature learning and selection (Ando& Zhang, 2005; Obozinski et al., 2006; Amit et al., 2007; Argyriou et al., 2008).”
Learning to Learn: Model Regression Networks for Easy Small Sample Learning (2016, background): “In particular, inter-class transfer [40] and crossgeneralization [7] are achieved by discovering shared feature representations: 1) captured by linear or nonlinear feature transformations [14, 63, 85, 31, 1, 94, 48], 2) obtained by feature selection [59, 60, 27] or regularization [37], 3) described ”
Tensor completion for estimating missing values in visual data (2010, background): “computer vision [20], machine learning [1, 2], and bioinformatics [21].”