Dictionary Learning from Ambiguously Labeled Data
Authors:
- Yi-Chen ChenYCYi-Chen Chen
- Vishal M. PatelVPVishal M. Patel
- Jaishanker K. PillaiJPJaishanker K. Pillai
- Rama ChellappaRCRama Chellappa
- P. Jonathon PhillipsPPP. Jonathon Phillips
- 1 June 2013
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 353-360
Abstract
We propose a novel dictionary-based learning method for ambiguously labeled multiclass classification, where each training sample has multiple labels and only one of them is the correct label. The dictionary learning problem is solved using an iterative alternating algorithm. At each iteration of the algorithm, two alternating steps are performed: a confidence update and a dictionary update. The confidence of each sample is defined as the probability distribution on its ambiguous labels. The dictionaries are updated using either soft (EM-based) or hard decision rules. Extensive evaluations on existing datasets demonstrate that the proposed method performs significantly better than state-of-the-art ambiguously labeled learning approaches.Keywords
This publication has 13 references indexed in Scilit:
- Rotation invariant simultaneous clustering and dictionary learning2012
- Task-Driven Dictionary LearningIEEE Transactions on Pattern Analysis and Machine Intelligence, 2011
- Dictionary learning and sparse coding for unsupervised clustering2010
- Sparse subspace clustering2009
- Learning from ambiguously labeled images2009
- Discriminative learned dictionaries for local image analysis2008
- Unsupervised Learning of Invariant Feature Hierarchies with Applications to Object Recognition2007
- Unsupervised Joint Alignment of Complex Images2007
- $rm K$-SVD: An Algorithm for Designing Overcomplete Dictionaries for Sparse RepresentationIEEE Transactions on Signal Processing, 2006
- The CMU pose, illumination, and expression databaseIEEE Transactions on Pattern Analysis and Machine Intelligence, 2003