AI Researchers Talk Up Benefits of GPUs for Deep Learning

With the ability to deliver TF/s to PF/s of performance even on nonlinear problems, deep-learning researchers who participated in the ImageNet competition are espousing the charms of GPU computing technology. At the European Conference on Computer Vision (ECCV), held last week in Zurich, teams from Adobe, U.C. Berkeley, the National University of Singapore, Oxford University and many others from around the world shared details on how GPUs helped them in the ImageNet competition. Given that training a neural network is a numerical optimization problem, it is not surprising that more performance equates to lower error on the training set as seen in the figure below. In addition, pruning heuristics can be applied to reduce parameter counts and increase generalization .

Deep-Learning error by year for ImageNet