This page contains whitepapers, articles and other data created by Numeric Insight. You are welcome to download and print the linked documents for your personal reference. Read the Terms of Use.
Your e-mail address is not required to download documents. No identifying information will be collected from you.
However, you are welcome to contact us to discuss the contents, or to suggest topics for articles yet to be posted here. Click on the links below to read/download material described below.
- A Gentle Introduction to Backpropagation is an article authored by Shashi Sathyanarayana which explains the algorithm which forms the backbone of most of deep learning.
- NMNIST.zip is a link to a compressed file containing a modified version of the famous MNIST database of handwritten digits. The modification consists of adding a small amount of random clutter to each of the original images. The purpose is to make classification and clustering problems using this data more challenging. Four uncompressed files are provided to serve as a direct drop-in replacement of the original MNIST dataset. A readme file within the compressed archive provides more information.
- Shapes_Data (zip) is a link to a compressed file containing the "shapes dataset". Each example is a 784-dimensional vector, which when reformatted as an image has the appearance of one of 3 shapes - triangle, square or pizza. A readme file within the compressed archive provides more information.