This WANN online demo interface is implemented with the Tensorflow.js API. It provides three experimental setups on which WANN domain adaptation algorithm can be tested with different hyper-parameters settings and compared to basic DA strategies.
Find more information on WANN algorithm and a complete Python implementation of the experiments on this GitHub Repository.

We consider here a synthetic experimental setup where source and target input instances are drawn uniformly on the [0, 1] interval. Source and target instances follow labeling functions of the form:

With - Amp. shift < < Amp. shift and given by Noise lvl or Tgt noise lvl (for target training data).
Neural networks with two layers of 100 neurons and ReLU activations are used to learn the task. To reduce computational time, a linear network is used as discrepancy hypothesis of WANN algorithm.

Parameters:

Experiments are conducted here on Kin-8xy which is a family of datasets synthetically generated from a realistic simulation of the forward kinematics of an 8 link all-revolute robot arm. The task consists in predicting the distance of the end-effector from a target. The task for each dataset has a specific degree of noise (moderate "m" or high "h") and linearity (fairly-linear "f", non-linear "n"). Each dataset is made of around 8000 row data and 8 features
Shallow networks of 100 neurons and a projecting constant given by Proj. Constant are used to learn the task.

Parameters:

Experiments are conducted here on the Superconductivity UCI dataset. The goal is to predict the critical temperature of superconductors based on features extracted from their chemical formula. The dataset contains around 20000 row data and 80 features. Four domains are defined by splitting along one input feature moderately correlated to the output.
Shallow networks of 100 neurons and a projecting constant given by Proj. Constant are used to learn the task.

Parameters: