Research map / Federated learning and privacy / Personalized and heterogeneous FL
TransFR: Transferable Federated Recommendation with Adapter Tuning on Pre-trained Language Models
Honglei Zhang, Zhiwei Li, Haoxuan Li, Xin Zhou, Jie Zhang, Yidong Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use textual item embeddings with adapter tuning instead of the best ID-based federated baseline FedDCSR | federated recommendation · Amazon cross-domain (average) | HR@10 transfer drop (Delta): 12.29% -> 4.33% | Improved |
| TransFR instead of FedDCSR on target-domain accuracy | federated recommendation · Amazon cross-domain (average) | target HR@10: 0.1939 -> 0.2182 | Improved |
| TransFR instead of FedDCSR in cold-start | federated recommendation · Amazon Music->Movies (cold-start) | HR@10: 0.1881 -> 0.2001 | Improved |
| add pre-training (vs random init) (ablation) | TransFR · Amazon Music->Books, Music->Movies, Books->Movies | HR@10 / NDCG@10 | Improved |
| add federated adapter-tuning (ablation) | TransFR · Amazon cross-domain | HR@10 / NDCG@10 | Improved |
| add post-adaptation personalization (ablation) | TransFR · Amazon cross-domain | HR@10 / NDCG@10 | Improved |
| add differential privacy noise (0.0 -> 0.3) (ablation) | TransFR · Amazon Music->Books (cold-start) | HR@10: 0.2310 -> 0.2234 | Hurt |
| replace DistBERT with Llama-2-7B (ablation) | TransFR · Amazon cross-domain | HR@10: 0.2182 -> 0.2241 | Improved |
| replace DistBERT with Llama-2-7B (ablation) | TransFR · Amazon cross-domain | memory (MB): 68 -> 14,336 | Hurt |
| increase adapter layers (ablation) | TransFR · Amazon transfers | HR@10 / NDCG@10 | Mixed |
| increase local epochs in post-adaptation personalization (ablation) | TransFR · Amazon transfers | HR@10 / NDCG@10 | Mixed |
| TransFR vs FedDCSR on source-domain accuracy | federated recommendation · Amazon Music->Books | source HR@10: 0.2417 -> 0.2386 | No clear effect |
Closest papers by meaning
- Federated Recommendation with Additive Personalization (ICLR 2024 · similarity 0.89)
- Dual-Personalizing Adapter for Federated Foundation Models (NeurIPS 2024 · similarity 0.88)
- Federated Graph Learning for Cross-Domain Recommendation (NeurIPS 2024 · similarity 0.87)
- Personalized Federated Collaborative Filtering: A Variational AutoEncoder Approach (AAAI 2025 · similarity 0.87)
- FLUX: Efficient Descriptor-Driven Clustered Federated Learning under Arbitrary Distribution Shifts (NeurIPS 2025 · similarity 0.87)
- FlowerTune: A Cross-Domain Benchmark for Federated Fine-Tuning of Large Language Models (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
