In this paper, we address the problem of domain adaptation in a regression setting, where only a few labeled samples are available in the source domain, and no labeled samples in the target domain. In addition, we consider that source data have different representations (multiple views). In this work, we investigate an original method to take advantage of different representations using a weighted combination of views. Besides, we use a co-training approach to include information from unlabeled instances by ensuring that models trained on different views make similar predictions. For this purpose, we introduce a novel formulation of the optimization objective for domain adaptation that relies on a discrepancy distance. Then, we develop an adversarial network domain adaptation algorithm adjusting weights given to each view, ensuring that those related to the target receive higher weights. Finally, we evaluate our method on different public datasets and compare it to other domain adaptation baselines to demonstrate the improvement for regression tasks.