This Google patent is about a training framework for performing machine learning on Extremely Large Datasets. It is looking like it is focusing on videos on Youtube. LinkedIn Shows Work on vision and video for the inventors of this patent.
The patent relates to the MapReduce-based training framework that exploits data parallelism and model parallelism to enable learning over a vast dataset.
In the last decade, a series of breakthroughs in machine learning and computer vision problems got attributed to the availability of Extremely Large Datasets. As the quality and quantity of datasets increased, so did the sophistication of models and their ability to accomplish more complex, high-level tasks such as:
Scene understanding
Pixel-level segmentation
Depth extraction
Visual-Question-Answering
Other image or video understanding tasks
However, for specific data modalities and learning scenarios, the size and number of available training examples can raise significant challenges, including, for example,…