We present MI-CRF, a conditional random field (CRF) model for multiple instance learning (MIL). MI-CRF models bags as nodes in a CRF with instances as their states. It combines discriminative unary instance classifiers and pairwise dissimilarity measures. We show that both forces improve the classification performance. Unlike other approaches, MI-CRF considers all bags jointly during training as well as during testing. This makes it possible to classify test bags in an imputation setup. The parameters of MI-CRF are learned using constraint generation. Furthermore, we show that MI-CRF can incorporate previous MIL algorithms to improve on their results. MI-CRF obtains competitive results on five standard MIL datasets.
%0 Conference Paper
%1 Deselaers2010
%A Deselaers, T.
%A Ferrari, V.
%B ICML
%D 2010
%K Learning classification ConditionalRandomField Optimization FeatureSelection
%T A Conditional Random Field for Multiple-Instance Learning
%U http://www.vision.ee.ethz.ch/publications/get_abstract.cgi?procs=730&mode=&lang=de
%X We present MI-CRF, a conditional random field (CRF) model for multiple instance learning (MIL). MI-CRF models bags as nodes in a CRF with instances as their states. It combines discriminative unary instance classifiers and pairwise dissimilarity measures. We show that both forces improve the classification performance. Unlike other approaches, MI-CRF considers all bags jointly during training as well as during testing. This makes it possible to classify test bags in an imputation setup. The parameters of MI-CRF are learned using constraint generation. Furthermore, we show that MI-CRF can incorporate previous MIL algorithms to improve on their results. MI-CRF obtains competitive results on five standard MIL datasets.
@inproceedings{Deselaers2010,
abstract = {We present MI-CRF, a conditional random field (CRF) model for multiple instance learning (MIL). MI-CRF models bags as nodes in a CRF with instances as their states. It combines discriminative unary instance classifiers and pairwise dissimilarity measures. We show that both forces improve the classification performance. Unlike other approaches, MI-CRF considers all bags jointly during training as well as during testing. This makes it possible to classify test bags in an imputation setup. The parameters of MI-CRF are learned using constraint generation. Furthermore, we show that MI-CRF can incorporate previous MIL algorithms to improve on their results. MI-CRF obtains competitive results on five standard MIL datasets.},
added-at = {2010-12-09T16:35:40.000+0100},
author = {Deselaers, T. and Ferrari, V.},
biburl = {https://www.bibsonomy.org/bibtex/29195f1b68359b53cb2a5b2bdb6b4a381/ipi_jn},
booktitle = {ICML},
file = {CRF Learning:E\:\\Literatursammlung\\Proceedings\\Sonstige\\Deselaers (2010) - A Conditional Random Field For Multiple-Instance Learning.pdf:PDF},
interhash = {8068a2781c0fe609b038989082de4c4d},
intrahash = {9195f1b68359b53cb2a5b2bdb6b4a381},
keywords = {Learning classification ConditionalRandomField Optimization FeatureSelection},
month = {June},
timestamp = {2010-12-09T16:35:40.000+0100},
title = {A Conditional Random Field for Multiple-Instance Learning},
url = {http://www.vision.ee.ethz.ch/publications/get_abstract.cgi?procs=730&mode=&lang=de},
year = 2010
}