Abstract: We propose a new method for domain adaptation that uses a gradient reversal layer (GRL) with instance weighting. Domain adaptation methods that use GRL have the latent problem of learning ...