Work

Transient objective convergence of spectrally weighted gradient descent

Year: 2026

Type: preprint

Abstract: AI-assisted, unrefereed candidate on ordinary-objective transient convergence for spectrally weighted single-gradient-direction methods. With smallest eigenvalue one, positive nondecreasing unbounded weights, fixed horizon K and relaxation 0<omega<=1, the worst-case ratio has leading deficit 2 omega K/kappa. Power weights admit sublogarithmic sharpness and a logarithmic interval of simultaneous failures of a strong two-step interpretation. The weaker interpretation remains true; no proved source theorem is refuted.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.24001733

Author: Anonymous

Language: English

Topic: Stochastic Gradient Optimization Techniques

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7225385941

Cites:

Cited by:

Related to: 10