‘…deep learning is less adept at interpreting video. Analyzing a video frame won’t reveal what’s happening unless that frame is compared with the ones that come before and after—a person holding a door may be opening it or closing it’
A group from MIT and IBM think they’ve found a way to adapt deep learning algorithms to analyze videos. The issues are:
- Context: is it in the air? on land? in water?
- Compute-intensity: can it be done without a big carbon footprint?
- Volume: 500 hours a minute, every day, is uploaded to YouTube as of May 2019, translating to 720,000 hours of new video posted daily.
- Potential uses: range from monitoring offensive material to selling ads.