Meta Proposes SlimPer Architecture for Personalized Recommendation

_reachsumit · x · 2026-07-15

Meta released SlimPer: Make Personalization Model Slim and Smart, proposing a new recommendation architecture.

The core idea of the paper is to reframe the ranking problem as an iterative refinement over a compact knowledge base, moving away from transformer-style sequential modeling. This solution has already been deployed on Instagram Reels and Feed.

Original post →

More from Research

Research channel →