Research map / Federated learning and privacy / Personalized and heterogeneous FL
Oblivionis: A Lightweight Learning and Unlearning Framework for Federated Large Language Models
Fuyao Zhang, Xinyu Yan, Tiantong Wu, Wenjie Li, Tianxiang Chen, Yang Cao, Ran Yan, Longtao Huang, Wei Yang Bryan Lim, Qiang Yang
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 FedAdagrad instead of FedAvg | SimNPO unlearning (Llama-3.2-1B) · TOFU Split99 | Forget Truth Ratio: 0.65 -> 0.74 | Improved |
| use FedAdagrad instead of FedAvg | SimNPO unlearning (Llama-3.2-1B) · TOFU Split99 | Model Utility: 0.46 -> 0.42 | Hurt |
| use FedAvgM instead of FedAvg | SimNPO unlearning (Llama-3.2-1B) · TOFU Split99 | Model Utility: 0.46 -> 1.8e-4 | Hurt |
| use FedProx instead of FedAvg | GradDiff unlearning (Llama-2-7B) · MUSE News | Utility Preserved: 0.34 -> 0.52 | Improved |
| use FedProx instead of FedAvg | GradDiff unlearning (Llama-2-7B) · MUSE News | No Verbatim Memorization: NVM 0.39 -> 0.52 | Hurt |
| scale base model from Llama-3.2-1B to 3B | FedAvg + Retrain · TOFU Split99 | Model Utility: 0.51 -> 0.59 | Improved |
| use federated learning instead of local training | GradAscent · TOFU Split99 (Llama-3.2-1B) | Model Utility: 0.00 -> 0.46 | Improved |
| use federated learning instead of local training | SimNPO · TOFU Split99 (Llama-3.2-1B) | Model Utility: 0.35 -> 0.46 | Improved |
| apply unlearning algorithms after federated fine-tuning | FedAvg (Llama-3.2-1B) · TOFU Split99 | Forget Truth Ratio: 0.49 -> 0.65 | Improved |
| apply unlearning after federated fine-tuning | FedAvg (Llama-3.2-1B) · TOFU Split99 | Model Utility: 0.50 -> 0.46 | Hurt |
Closest papers by meaning
- FlowerTune: A Cross-Domain Benchmark for Federated Fine-Tuning of Large Language Models (NeurIPS 2025 · similarity 0.91)
- FedRW: Efficient Privacy-Preserving Data Reweighting for Enhancing Federated Learning of Language Models (NeurIPS 2025 · similarity 0.90)
- Splitting with Importance-aware Updating for Heterogeneous Federated Learning with Large Language Models (ICML 2025 · similarity 0.89)
- Differentially Private Federated Low Rank Adaptation Beyond Fixed-Matrix (NeurIPS 2025 · similarity 0.89)
- Unified Parameter-Efficient Unlearning for LLMs (ICLR 2025 · similarity 0.89)
- OpenUnlearning: Accelerating LLM Unlearning via Unified Benchmarking of Methods and Metrics (NeurIPS 2025 · similarity 0.89)
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.
