Training of a Neural Network for learning a model of joint transformations

The data contains joint angle data from walking stick insects. It provides matching sets of two neighboring legs that lead these two legs towards the same position in space. During climbing through twigs insects exploit the information on footholds between these neighboring legs. This requires transformations that mediate between the legs, in order to guide one leg to the position of the other leg. Here, different neural network architectures are learned and compared. The neural networks are realized in the keras framework. The python scripts compare different neural network architectures (different size of hidden layer or introducing skip connections) evaluating the Mean Squared error. Recent versions can be found at: https://github.com/malteschilling/joint_mappings_nn