Skin lesion segmentation (SLS) in dermoscopic images is a crucial task for\nautomated diagnosis of melanoma. In this paper, we present a robust deep\nlearning SLS model, so-called SLSDeep, which is represented as an\nencoder-decoder network. The encoder network is constructed by dilated residual\nlayers, in turn, a pyramid pooling network followed by three convolution layers\nis used for the decoder. Unlike the traditional methods employing a\ncross-entropy loss, we investigated a loss function by combining both Negative\nLog Likelihood (NLL) and End Point Error (EPE) to accurately segment the\nmelanoma regions with sharp boundaries. The robustness of the proposed model\nwas evaluated on two public databases: ISBI 2016 and 2017 for skin lesion\nanalysis towards melanoma detection challenge. The proposed model outperforms\nthe state-of-the-art methods in terms of segmentation accuracy. Moreover, it is\ncapable to segment more than $100$ images of size 384x384 per second on a\nrecent GPU.\n