A sequential dynamic multi-class model and recursive filtering by variational bayesian methods

Authors:
Xiangyun Qing;Xingyu Wang
Affiliations:
School of Information Science and Engineering, East China University of Science and Technology, China;School of Information Science and Engineering, East China University of Science and Technology, China
Venue:
PAKDD'11 Proceedings of the 15th Pacific-Asia conference on Advances in knowledge discovery and data mining - Volume Part I
Year:
2011

Citing 7
Cited 0

Mining time-changing data streams

Proceedings of the seventh ACM SIGKDD international conference on Knowledge discovery and data mining
Variational Bayesian multinomial probit regression with Gaussian process priors

Neural Computation
Adaptive Learning Rate for Online Linear Discriminant Classifiers

SSPR & SPR '08 Proceedings of the 2008 Joint IAPR International Workshop on Structural, Syntactic, and Statistical Pattern Recognition
Adapted One-versus-All Decision Trees for Data Stream Classification

IEEE Transactions on Knowledge and Data Engineering
2009 Special Issue: Adaptive classification for Brain Computer Interface systems using Sequential Monte Carlo sampling

Neural Networks
Sequential non-stationary dynamic classification with sparse feedback

Pattern Recognition
A comparison of methods for multiclass support vector machines

IEEE Transactions on Neural Networks

Quantified Score

Hi-index	0.00

Visualization

Abstract

Adaptive classification evolving over time is an important learning task that arises in many applications. In this paper, a sequential dynamic multi-class model (SDMM) is proposed for representing the multi-class adaptive learning task, which is based on the polychotomous response model and dynamic logistic regression. Multiple state chains in the SDMM are coupled by the observable labels and feature vectors. Each state chain is modeled as a first-order Markov process with timevarying covariance parameters for characterizing the non-stationary generating process of sequential labels. Augmented auxiliary variables are introduced for developing efficient inference procedures according to the popular data augmentation strategy. Variational Bayesian methods are applied to estimate the dynamic state variables and augmented auxiliary variables recursively. According to the results of recursive filtering procedures using mean-field approximation forms, one-step-ahead predicted probabilities are calculated by marginalizing the state variables. Experiment results based on both synthetic and real data show that the proposed model significantly outperforms the non-sequential static methods for the multi-class adaptive learning problems with missing labels. Encouraging results have been obtained by comprising well-known multi-class data stream algorithms.