DRJ's Fall 2018

Conference & Exhibit

Attend The #1 BC/DR Event!

Summer Journal

Volume 31, Issue 2

Full Contents Now Available!

Tuesday, 15 May 2018 15:57

Data Storage, AI, and IO Patterns

AI is one of the hot new topics in computing and with good reason. New techniques in Deep Learning (DL) involving Neural Networks (NN), has the ability to create NN's that achieve better than human accuracy on some problems. Image recognition is an example of how DL models can achieve better than human accuracy in identifying objects in images (object detection and classification).

An example of this is the Imagenet competition. Since 2010 the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) has been used as a gauge of the improvement in image recognition. In 2011, the best error rate was around 25% (the tool could correctly identify 75% of the images outside of the training data set). In 2012, a Deep Neural network (DNN) had an error rate of 16%. In the next few years the error percentage dropped to the single digits. In 2017 29 of the 36 competing teams got less than 5% wrong which is typically better than a human.