Frans M. J. Willems

Learn More
We describe a sequential universal data compression procedure for binary tree sources that performs the “double mixture.” Using a context tree, this method weights in an efficient recursive way the coding distributions corresponding to all bounded memory tree sources, and achieves a desirable coding distribution for tree sources with an unknown model and(More)
First we modify the basic (binary) context-tree weighting method such that the past symbols x 1?D ; x 2?D ; ; x 0 are not needed by the encoder and the decoder. Then we describe how to make the context tree depth D innnite, which results in optimal redundancy behavior for all tree sources, while the number of records in the context tree is not larger than(More)
This paper discusses a switching method which can be used to combine two sequential universal source coding algorithms. The switching method treats these two algorithms as black-boxes and can only use their estimates of the probability distributions for the consecutive symbols of the source sequence. Three weighting algorithms based on this switching method(More)
This paper addresses privacy leakage in biometric secrecy systems. Four settings are investigated. The first one is the standard Ahlswede-Csiszar secret-generation setting in which two terminals observe two correlated sequences. They form a common secret by interchanging a public message. This message should only contain a negligible amount of information(More)