Researchers at Tel Aviv University in Israel and Cornell University have developed neural network-based detection algorithms that can read what movies people who use streaming services are viewing, even when they cannot directly monitor the stream of bits delivering them, or access the device on which users are watching them.
The technique identifies the video "fingerprint" of streaming content mined from a pattern of information "bursts" derived from the streaming service's application of variable bit-rate compression.
The researchers say they can determine the video fingerprint with up to 99% accuracy by comparing a pre-assembled library of prints the hacker has compiled from videos whose viewers they might want to surveil.
From The Economist
View Full Article