Backpropagation Algorithm in neural net

ssv112

Hello,
I am building a classification model using Neural Network operator. I want to know which backpropagation algorithm is used i this operater for weight update and reducing the error ( while training) . Like in matlab we have trainbfg , traincgf...
What is used here? and is this algorithm takes global minima or local minima  ( for minimum error)

varunm1

Are you asking about the Deep Learning (H20 based) operator or the deep learning extension? If it is  about the deep learning operator (H20) it is a feed forward network with SGD based back propagation. The difference between a neural network operator and  deep learning operator is it is highly customizable with different activations and  regularizations. For RNN and CNN, we have deep learning extension which needs to be installed from marketplace.

MarlaBot

Hi @ssv112 - this is MarlaBot. I found these great videos on our RapidMiner Academy that you may find helpful:

Instructional Video: Sampling & Weighting demo (Viewing time: ~7m)

Instructional Video: Sampling & Weighting intro (Viewing time: ~5m)

Instructional Video: Introduction to the k-NN Algorithm (Viewing time: ~4m)

5 Min with Ingo Video: Understanding K Nearest Neighbors Algorithms (Viewing Time: ~6m)

Please LIKE my comment if it helps! 👇

MarlaBot

varunm1

Hello @ssv112

After some time looking at the RM neural net code, I understood that it is using the JOONE engine (Java-based) with momentum backpropagation. Coming to the question 2, the optimization is local optimization (converge to optimal local minima) based on momentum.

@IngoRM anything to add here?

Hope this helps

ssv112

Thanks guys! There was one more issue.. deeplearning is having feed forward backpropagation neural network.. Isnt this supposed to b having recurrent neural network like having feedback?? I feel its a normal ANN with more neurons in hidden layer only! @IngoRM @varunm1