Research map / Federated learning and privacy / Differential privacy
Learning To Generate Image Embeddings With User-Level Differential Privacy
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 DP-FedEmb instead of DP-FedAvg | User-level DP embedding training · DigiFace test | Recall@FAR: 13.38±0.27 -> 19.76±0.43 | Helped |
| Use DP-FedEmb instead of DP-FedAvg | User-level DP embedding training · EMNIST (Approx) | Recall@FAR: 9.78±0.44 -> 10.64±0.5 | Helped |
| Use DP-FedEmb instead of DP-FedAvg | User-level DP embedding training · GLD (AllPair) | Recall@FAR: 26.27±0.17 -> 27.07±0.04 | Helped |
| Use DP-FedEmb instead of DP-FedAvg | User-level DP embedding training · iNat (AllPair) | Recall@FAR: 29.6±0.65 -> 40.93±0.94 | Helped |
| Add ImageNet pretraining (ablation) | DP-FedEmb MobileNetV2 · DigiFace10K, 400 rounds | Recall@FAR: 57.12% -> 65.19% | Helped |
| Add ImageNet pretraining (ablation) | DP-FedEmb MobileNetV2 · DigiFace10K, 800 rounds | Recall@FAR: 62.76% -> 68.86% | Helped |
| Pretrain on the preprocessed GLD dataset rather than using no pretraining (ablation) | DP-FedEmb · DigiFace10K | Recall@FAR | Hurt |
| Freeze intermediate convolutional blocks (ablation) | DP-FedEmb ResNet-50 · DigiFace10K, small-medium noise | Recall@FAR | Hurt |
| Freeze intermediate convolutional blocks (ablation) | DP-FedEmb ResNet-50 · DigiFace10K, noise 0.4 | Recall@FAR: 0.4 | Helped |
| Add noise multiplier 0.02 (ablation) | DP-FedEmb-r800 MobileNetV2 · DigiFace10K | Recall@FAR: 68.86% -> 67.56% | Hurt |
| Add noise multiplier 0.02 (ablation) | DP-FedEmb-r800 ResNet-50 · DigiFace10K | Recall@FAR: 79.09% -> 72.6% | Hurt |
| Replace local fine-tuning with head reconstruction (ablation) | DP-FedEmb · DigiFace10K | Recall@FAR | No clear effect |
Closest papers by meaning
- Private Model Personalization Revisited (ICML 2025 · similarity 0.88)
- Differentially Private Representation Learning via Image Captioning (ICML 2024 · similarity 0.87)
- Theoretically Unmasking Inference Attacks Against LDP-Protected Clients in Federated Vision Models (ICML 2025 · similarity 0.87)
- Enabling Differentially Private Federated Learning for Speech Recognition: Benchmarks, Adaptive Optimizers, and Gradient Clipping (NeurIPS 2025 · similarity 0.86)
- Privacy-Preserving Personalized Federated Prompt Learning for Multimodal Large Language Models (ICLR 2025 · similarity 0.86)
- Clients Collaborate: Flexible Differentially Private Federated Learning with Guaranteed Improvement of Utility-Privacy Trade-off (ICML 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.