Sparse Bayesian Simplified Models for ECG Compression



Three algorithms are outlined, each of which with interesting features,
for building a model for ECG Compression. Recent work in machine learning
has focused in models, such as Support Vector Machines (SVMs) that automatically
control generalization and parameterization as part of the overall optimization
process. Its easiness of structural model finding made them surpass Multiple
Feed-Forward Networks (MFFs) for estimation problems. Ultimately Relevance
Vector Machines (RVMs) become very popular since they maintain these advantages
while making predictions of the probability distributions of model parameters
rather than point estimates on the test data set. Prediction results of the
three learning machines in an ECG model compression problem are assessed using
conventional metrics. Results show that sparsity is fully explored in RVMs
making them candidates of choice for real time applications.


Biomedical Engineering

Related Project

IST FP6 MyHeart


International Conference on Natural Computation (ICNC\'06), September 2006

Cited by

No citations found