Recurrent Neural Networks with Weighting Loss for Early Prediction of Hand Movements

We propose in this work an approach for earlyprediction of hand movements using recurrent neural networks(RNNs) and a novel weighting loss. The proposed loss functionleverages the outputs of an RNN at different time steps andweights their contributions to the final loss linearly over timesteps. Additionally, a sample weighting scheme also constitutesa part of the weighting loss to deal with the scarcity of thesamples where a change of hand movements takes place. Theexperiments conducted with the Ninapro database reveal thatour proposed approach not only improves the performance inthe early prediction task but also obtains state of the art resultsin classification of hand movements. These results are especiallypromising for the amputees.