Research map / Optimization / Optimizers

Calibrated Target Noise Recovers Curvature from the Gradients

NeurIPS 2026 · Optimization › Optimizers · approach: Natural gradient and Fisher approximations

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes calibrated target-perturbation noise as a simple scheme for recovering curvature information from gradients alone, motivated by settings with only aggregate mini-batch gradients.

Problem

Estimating curvature (Hessian-related) information from gradients alone when only aggregate mini-batch gradients are available, which is needed for preconditioning, sampling, and early stopping.

Approach

A simple target-perturbation scheme with calibrated noise on the regression targets is claimed to suffice for recovering curvature from gradients; the abstract is truncated so the exact mechanism is not stated.

Surprising findings

• The abstract claims that a simple target-perturbation scheme suffices to recover curvature from gradients, which the authors describe as perhaps surprising.

Closest papers by meaning

• Meta-Curvature (NeurIPS 2019 · similarity 0.89)
• Global curvature for second-order optimization of neural networks (ICML 2025 · similarity 0.87)
• Improving Computational Complexity in Statistical Models with Local Curvature Information (ICML 2024 · similarity 0.87)
• Stepping on the Edge: Curvature Aware Learning Rate Tuners (NeurIPS 2024 · similarity 0.86)
• Deconstructing the Goldilocks Zone of Neural Network Initialization (ICML 2024 · similarity 0.86)
• Curvature Tuning: Provable Training-free Model Steering From a Single Parameter (NeurIPS 2025 · similarity 0.86)