Galaxies exhibit a wide variety of morphologies which contain valuable information about their star formation histories. Having large samples of morphologically classified galaxies is fundamental to understand their formation and evolution. Deep learning algorithms have proven to be extremely successful for morphological classification of galaxies. However, these algorithms rely on large training sets (~5000 galaxies) of pre-labelled galaxies belonging to the same domain (e.g., dataset or redshift). I will show two alternatives for classifying galaxies from the Dark Energy Survey, where the training sample is limited. 1) Transfer of knowledge: we test the performance of Deep Learning models, trained with SDSS data, on Dark Energy survey images. After a fast domain adaptation step, machines can quickly adapt to new instrument characteristics (e.g., PSF, seeing, depth), reducing by almost one order of magnitude the necessary training sample for morphological classification. 2) Simulations: current morphological catalogues are limited to very bright observed magnitudes (mr < 18). We use galaxies with known morphological classifications and simulate them at higher redshifts. Using a combination of real and simulated images as training sample, we demonstrate that machines can recover features hidden to the human eye. Thanks to this approach, we are able to push our morphological classification to fainter magnitudes (mr < 21). We have publicly released the largest morphological catalogue up to date, containing ~25 million galaxies.