hi David, from this paragraph, it seems to be saying that FFM is more granular due to the multiple…

Comparing FFM versus FM, FFM learns multiple latent vector for each feature while FM learns a latent vector for each feature. One can interpret the former as trying to represent the interactions in a more granular-level. As such, the number of latent features k needed to represent such granular interactions is lesser i.e. k in FFM << k in FM.
An Intuitive Explanation of Field Aware Factorization Machines
320
2
David Chong
Kenneth Leung
·Follow
Jul 29, 2021
--
hi David, from this paragraph, it seems to be saying that FFM is more granular due to the multiple latent vectors per feature. However, in the last sentence, it seems to suggest the opposite, where you mentioned that the k in FFM is much lower (<<) than the k in FM. Just to confirm that this is correct?
--
--
Written by Kenneth Leung9.5K Followers
·182 Following
Senior Data Scientist at Boston Consulting Group | Top Tech Author | 2M+ reads on Medium | linkedin.com/in/kennethleungty | github.com/kennethleungty
No responses yet
Help
Status
About
Careers
Press
Blog
Privacy
Terms
Text to speech
Teams