Embedding and ranking for images, entities, items and text

Jason Weston

We describe several recent results of applying embedding and ranking techniques to the real-world problems of image search and annotation, relation extraction, and music and video recommendation (Google Music and YouTube). Key issues include modifying the algorithms to increase their capacity, and making them scale to the size of the task (particularly at test time). This is joint work with Samy Bengio, Maya Gupta, Aurelien Lucchi, Antoine Bordes, Nicolas Usunier, Oksana Yakhnenko, Ron Weiss, Hector Yee and Ameesh Makadia.

Presentation (PDF File)

Back to Stochastic Gradient Methods