Home
Flash News
New CTCL framework generates privacy-preserving synthetic data using a 140M-parameter model, avoiding costly billion-scale LLM fine-tuning. It matches topic distributions, scales efficiently, and outperforms baselines in diverse datasets under strong privacy guarantees, as shown at ICML 2025.
New CTCL framework generates privacy-preserving synthetic data using a 140M-parameter model, avoiding costly billion-scale LLM fine-tuning. It matches topic distributions, scales efficiently, and outperforms baselines in diverse datasets under strong privacy guarantees, as shown at ICML 2025.
Comments (0)
0/300
