incorrect output values using Keras dense layers

Hi, I am new to Knime. I built a neuralnetwork to predict numeric output values (full precision, one output) for the given input (3 inputs). When using Keras Input, dense and output layers, with ReLU activation and Adam optimizer, the output values are all zeroes. However, using RProp MLP learner node and multilayer perceptron predictor nodes, I was able to successfully train, test and get expected results. I would appreciate some tips to resolve theproblem using Keras layers.