Open-source research
FDRec
Frequency-decoupled knowledge distillation for reducing multimodal recommendation cost while preserving useful cross-modal signals.
- Year
- 2025
- Status
- active
- Focus
- Python · Knowledge Distillation · Recommendation
What it does
FDRec separates multimodal representations into low- and high-frequency components before distillation. Each component receives a loss suited to its information characteristics, with additional feature normalization and alignment.
The released implementation is designed to transfer multimodal knowledge into more efficient recommendation models while reducing inference overhead.
Evidence and scope
This page is a concise guide to the work. Use the primary paper for the experimental setup, quantitative results, limitations, and formal claims; use the repository for the implementation and current reproduction instructions.