Although NNs were proposed in the 1940s and DNNs in 1960s, the first practical application employing multiple digital neurons appeared in 1990 with the LeNet network for handwritten digit recognition. The Deep Learning (DL) successes of the 2010s are believed to be under the confluence of three main factors: 1. the new algorithmic advances that have improved application accuracy significantly and broadened applicable domains; 2. the availability of huge amount of data to train NNs; 3. the availability of enough computing capacity. Many DNN models have been developed over the past two decades [Deshpande 2017] [Kalray 2017] [Sze 2017] . Each of these models has a different network architecture in terms of number of layers, layer types, layer shapes and connections between layers. In Table 1 we present a timeline of some iconic computer vision models over the past years. Some of them will be presented along with their performance in a well-known computer vision challenge, the ImageNet Large-Scale Visual Recognition Challenge (ILSVRC) [Russakovsky 2015].

Return to Contemt