Extracting information from fine-tuned weights

By Nan Chen, Hayden S. Helm, Y. Park, Carey E. Priebe, Soledad Villar

Subfields: Representation & Self-Supervised Learning, Privacy & Security

Citations in the corpus

• Black-box model classification under the discriminative factorization (2026, unknown): “Low-dimensional representations of generative models can be constructed from internal activations [29, 21, 17, 10] and model weights [2, 7].”
• Query-efficient model evaluation using cached responses (2026, unknown): “Some approaches construct representations by comparing internal activations across models (Duderstadt et al., 2023; Huh al., 2024; Horwitz et al., 2025) or by comparing model weights directly (Chen et al., 2025).”