Google's New Compression is More Efficient than JPEGs

By: Joey Haar - Aug 24, 2016
References: & thenextweb
Researchers for Google at Cornell University have created a compression algorithm that sounds as though it's straight out of HBO's comedy series 'Silicon Valley.' The compression algorithm, which was announced in a recently released paper, uses a neural network and a "middle-out" system to compress images more effectively than the standard JPEG compression rate.

To understand the new compression algorithm, it's first essential to understand neural networks like Google's open source 'TensorFlow.' Neural networks are artificial intelligence systems that function in a similar way to the human brain, learning from patterns based on the input they receive.

For the new compression algorithm, the researchers fed TensorFlow six million images and selected the 100 least effective compression sections in each of these. It then had TensorFlow compress these difficult sections, giving it a sort of trial by fire. The outcome was a more efficient compression algorithm.