TY - GEN
T1 - Combining the advice of experts with randomized boosting for robust pattern recognition
AU - Peng, Jing
AU - Seetharaman, Guna
PY - 2013
Y1 - 2013
N2 - We have developed an algorithm, called ShareBoost, for combining mulitple classifiers from multiple information sources. The algorithm offer a number of advantages, such as increased confidence in decision-making, resulting from combined complementary data, good performance against noise, and the ability to exploit interplay between sensor subspaces.We have also developed a randomized version of ShareBoost, called rShare-Boost, by casting ShareBoost within an adversarial multi-armed bandit framework. This in turn allows us to show rShareBoost is efficient and convergent. Both algorithms have shown promise in a number of applications. The hallmark of these algorithms is a set of strategies for mining and exploiting the most informative sensor sources for a given situation. These strategies are computations performed by the algorithms. In this paper, we propose to consider strategies as advice given to an algorithm by 'experts' or 'Oracle.' In the context of pattern recognition, there can be several pattern recognition strategies. Each strategy makes different assumptions regarding the fidelity of each sensor source and uses different data to arrive at its estimates. Each strategy may place different trust in a sensor at different times, and each may be better in different situations. In this paper, we introduce a novel algorithm for combining the advice of the experts to achieve robust pattern recognition performance. We show that with high probability the algorithm seeks out the advice of the experts from decision relevant information sources for making optimal prediction. Finally, we provide experimental results using face and infrared image data that corroborate our theoretical analysis.
AB - We have developed an algorithm, called ShareBoost, for combining mulitple classifiers from multiple information sources. The algorithm offer a number of advantages, such as increased confidence in decision-making, resulting from combined complementary data, good performance against noise, and the ability to exploit interplay between sensor subspaces.We have also developed a randomized version of ShareBoost, called rShare-Boost, by casting ShareBoost within an adversarial multi-armed bandit framework. This in turn allows us to show rShareBoost is efficient and convergent. Both algorithms have shown promise in a number of applications. The hallmark of these algorithms is a set of strategies for mining and exploiting the most informative sensor sources for a given situation. These strategies are computations performed by the algorithms. In this paper, we propose to consider strategies as advice given to an algorithm by 'experts' or 'Oracle.' In the context of pattern recognition, there can be several pattern recognition strategies. Each strategy makes different assumptions regarding the fidelity of each sensor source and uses different data to arrive at its estimates. Each strategy may place different trust in a sensor at different times, and each may be better in different situations. In this paper, we introduce a novel algorithm for combining the advice of the experts to achieve robust pattern recognition performance. We show that with high probability the algorithm seeks out the advice of the experts from decision relevant information sources for making optimal prediction. Finally, we provide experimental results using face and infrared image data that corroborate our theoretical analysis.
UR - http://www.scopus.com/inward/record.url?scp=84898401025&partnerID=8YFLogxK
U2 - 10.1109/AIPR.2013.6749332
DO - 10.1109/AIPR.2013.6749332
M3 - Conference contribution
AN - SCOPUS:84898401025
SN - 9781479925407
T3 - Proceedings - Applied Imagery Pattern Recognition Workshop
BT - 2013 IEEE Applied Imagery Pattern Recognition Workshop
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2013 IEEE Applied Imagery Pattern Recognition Workshop: Sensing for Control and Augmentation, AIPR 2013
Y2 - 23 October 2013 through 25 October 2013
ER -