Algorithmic progress means that less compute and data are required to achieve a given level of performance

Amount of compute and number of data points required to achieve 80.9% accuracy on an image recognition test

ResNeXt-101 computer vision system on the ImageNet benchmark. Compute is measured in FLOPs (floating-point operations). Data is measured in the number of images in the training set.
Chart: Will Henshall for TIME Source: Epoch
TIME