My Google patent!
A happy surprise several years later…
A few years ago, I wrote about my PhD internship experience at Google, where I worked on machine learning for video compression. It was a summer in which I basically had no life and worked the entire 3 months, but it eventually paid off with the research turning into a paper and oral presentation at ICASSP 2023 in Rhodes, Greece.
In a fun and unexpected turn of events, I just discovered that the core research from that internship was also filed as a patent by Google, and it has since been published.
The patent is titled “Learned transforms for coding” with the publication number WO2024081009A1.
It covers the nonlinear transform model I developed, which uses a multi-rate adaptive autoencoder and hyperprior to improve compression efficiency on video prediction residuals.
