60 likes | 195 Views
Datamining & Neural Networks . Session 3 PCA & ARD. Regularization on cost-function : E D + E W e.g E W =1/2*||w|| 2 three levels of inference Infer model parameters w ; , fixed Infer , ; given the model params comparison of different models
E N D
Datamining & Neural Networks Session 3 PCA & ARD
Regularization on cost-function : ED + EW • e.g EW=1/2*||w||2 • three levels of inference • Infer model parameters w ; , fixed • Infer , ; given the model params • comparison of different models • Evidence-based framework of MacKay for estimation of hyperparameters : interpretation of • as 1/2W • as 1/2D • approx. of evidence factor yields update formula’s for hyperparameters P(|D)~P(D|) * P()
ARD • one hyperparam per input of your NN • Goal: detection of most relevant inputs in the Bayesian framework 1 2 N N 3 Costfunction = ED + 1(wi12)+2(wi22)+…
ARD in practice • Define NN model • Optimize weights • Optimize , , -- given weights • Cycle back to 2.
Exercise 2 • Demev1 (Evidence-based framework) • Demard (ARD) • Ionosphere dataset: 33 inputs, appr. 350 datapoints. • make classification NN (cfr. also session 2, ex. 3) • perform ARD, rank inputs • select subset of inputs based on ’s • make new NN with this subset of inputs • compare performance with first NN