On the Learnability and Design of Output Codes for Multiclass Problems

By Koby Crammer, Yoram Singer. 2002

788 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computational Theory and Mathematics

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 Computational Theory and Mathematics

Citations in the corpus

Stronger data poisoning attacks break data sanitization defenses (2018, methodology): “a We used the multi-class SVM formulation in Crammer and Singer (2002) with no explicit regularization for the 10-class MNIST dataset.”
Latent Embeddings for Zero-Shot Classification (2016, methodology): “The SGD algorithm for multiclass loss of Crammer and Singer [4] requires at each iteration a full pass over all the classes to search for the maximum violating class.”
Regularization Techniques for Learning with Matrices (2009, background): “Several examples are multi-class categorization (e.g. Crammer and Singer [2000]), multi-task and multi-view learning (e.g. Cavallanti et al. [2008], Agarwal et al. [2008]), and online PCA [Warmuth and Kuzmin, 2006].”
Support vector machine-based expert system for reliable heartbeat recognition (2004, methodology): “Although SVM separates the data only into two classes, the recognition of more classes is straightforward by applying either “one against one” or “one against all” methods [18], [19].”
Designing Category-Level Attributes for Discriminative Visual Recognition (2013, result): “The above framework is motivated by two previous studies: learning attributes based on category-attribute matrix [13], and Error Correcting Output Code (ECOC) [1,5].”
Spectral Embedded Clustering: A Framework for In-Sample and Out-of-Sample Spectral Clustering (2011, background): “However, as discussed in [30], the prediction performance highly depends on the design of error correcting output codes, which is a nontrivial task.”
Supervised Machine Learning: A Review of Classification Techniques (2007, positive): “SVMs are binary algorithm, thus we made use of error-correcting output coding (ECOC), or, in short, the output coding approach, to reduce a multi-class problem to a set of multiple binary classification problems (Crammer & Singer, 2002).”
Tighter Bounds for Structured Estimation (2008, methodology): “In this experiment, we investigate the performance of convex and ramp loss versions of the WinnerTakes-All multiclass classification [1] when the training data is noisy.”
Efficient projections onto the l1-ball for learning in high dimensions (2008, methodology): “Crammer and Singer (2002) rediscovered a similar projection algorithm as a tool for solving the dual of multiclass SVM.”
Leveraging Structure from Motion to Learn Discriminative Codebooks for Scalable Landmark Classification (2013, methodology): “We tried the SVM formulation proposed in [4], also used by [16], but did not see much difference in performance.”
Beyond One-hot Encoding: lower dimensional target embedding (2018, background): “In [9], it is also shown Optimal continuous ECOCs can be found by gradient descent.”
Machine learning: a review of classification and combining techniques (2006, methodology): “The SMO algorithm is a binary algorithm, thus we made use of error-correcting output coding (ECOC), or, in short, the output coding approach, to reduce a multi-class problem to a set of multiple binary classification problems (Crammer and Singer 2002).”
A comparative study of feature selection and multiclass classification methods for tissue classification based on gene expression (2004, background): “For example, Crammer and Singer (2000) study various strategies for building code-words for error-correcting output coding, and ﬁnd no clear winner.”
A comparison of methods for multiclass support vector machines (2002, methodology): “In [7], an algorithm was proposed for (18) while in [8], a simple iterative approach was used.”
A Hybrid Loss for Multiclass and Structured Prediction (2014, background): “This is intractable for labels where the possible assignments grow exponentially.”
Modeling transition patterns between events for temporal human action segmentation and classification (2015, methodology): “1) Training Segment-SVM with Max Margin Optimization: We learn discriminative patterns of each peak and transition segments using a multi-class SVM [6] similar to [9].”
An Introduction to Boosting and Leveraging (2002, neutral): “[76, 158, 62, 18, 31, 3, 155, 15, 44, 52, 82, 164, 66, 38, 137, 147, 16, 80, 185, 100, 14].”