Interesting read from Youtube Recommendations team (https://lnkd.in/e-k4YqN). They have used shallow tower (Wide and Deep Neural Networks) to remove position bias and introduced a Mixture of Experts to optimize for multiple objectives.
Another interesting variation was to use PointWise Similarity for the Learn to Rank Re-Ranker for scalability reasons.
It was also interesting to know that they tuned the weights of multiple experts manually.
What are your thoughts?
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment