Preprint · 2026 · Open access

Noisy Expected Improvement Forgets the Objective

Anonymous

Zenodo (CERN European Organization for Nuclear Research) · Published 11 August 2026 · DOI 10.5281/zenodo.21881512

Open full textCite this paper

In brief

“Noisy Expected Improvement Forgets the Objective” is a 2026 preprint by Anonymous, published in Zenodo (CERN European Organization for Nuclear Research). OpenAlex counts 0 citations as of 19 November 2026. It is open access under a CC BY licence: the full text is free to read at Zenodo (CERN European Organization for Nuclear Research).

Citations
0
Authors
1

Abstract

Expected improvement (EI) is commonly applied to noisy Bayesian optimization by inserting the best observed noisy value into the analytic acquisition formula. We characterize the resulting policy under a well-specified continuous Gaussian process prior and Gaussian observation noise. The best observation diverges at the Gaussian extreme-value scale, while all posterior means remain bounded almost surely. A deterministic comparison then shows that EI selects a point whose posterior standard deviation is within 1-O(1/sqrtlog t) of the maximum. Thus noisy best-observation EI becomes maximum-variance design and asymptotically forgets the objective. This has sharply different implications for bandit and optimization metrics. On any finite domain with a strictly positive-definite prior covariance, every fixed-quality approximate EI policy has linear cumulative regret. Exact EI allocates every point frequency 1/m. Nevertheless, the posterior-mean recommendation is eventually optimal. On compact domains, we prove the proper simple-regret guarantee O(sqrtgammaTlog T/T) almost surely, where gammaT is maximum information gain, and give a uniform high-probability version after an explicit burn-in. We also show that the raw noisy error used in recent analyses tends to -infty, so its convergence upper bounds do not certify optimization. These results resolve the long-run behavior of a standard noisy EI heuristic and separate observation extremes, cumulative regret, and recommendation quality.

Cite this paper

Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.

APA 7

Anonymous. (2026). Noisy Expected Improvement Forgets the Objective. Zenodo (CERN European Organization for Nuclear Research). https://doi.org/10.5281/zenodo.21881512

MLA 9

Anonymous. “Noisy Expected Improvement Forgets the Objective.” Zenodo (CERN European Organization for Nuclear Research), 2026. https://doi.org/10.5281/zenodo.21881512.

Chicago

Anonymous. 2026. “Noisy Expected Improvement Forgets the Objective.” Zenodo (CERN European Organization for Nuclear Research). https://doi.org/10.5281/zenodo.21881512.

Harvard

Anonymous (2026) ‘Noisy Expected Improvement Forgets the Objective’, Zenodo (CERN European Organization for Nuclear Research). doi:10.5281/zenodo.21881512.

IEEE

Anonymous, “Noisy Expected Improvement Forgets the Objective,” Zenodo (CERN European Organization for Nuclear Research), 2026, doi: 10.5281/zenodo.21881512.

BibTeX

@misc{anonymous2026noisy,
  title     = {{Noisy Expected Improvement Forgets the Objective}},
  author    = {Anonymous},
  howpublished = {Zenodo (CERN European Organization for Nuclear Research)},
  year      = {2026},
  doi       = {10.5281/zenodo.21881512},
  url       = {https://doi.org/10.5281/zenodo.21881512}
}

Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)

Topic and keywords

Social Sciences › Decision Sciences › Management Science and Operations Research › Advanced Bandit Algorithms Research

  • Gaussian process priors
  • measurement noise
  • regret analysis
  • simple regret
  • information gain
  • burn-in

Work with this paper

Most-cited papers in Decision Sciences

All highly cited Decision Sciences papers

Questions about this paper

How do I cite this paper in APA style?

Anonymous. (2026). Noisy Expected Improvement Forgets the Objective. Zenodo (CERN European Organization for Nuclear Research). https://doi.org/10.5281/zenodo.21881512

How many times has this paper been cited?

0 times, according to OpenAlex on 19 November 2026. Google Scholar and Scopus count different sets of sources, so their numbers differ.

Is this paper free to read?

Yes. It is open access at Zenodo (CERN European Organization for Nuclear Research): sign in and open the full text from this page.

Who wrote this paper?

Anonymous.

What is the DOI of this paper?

10.5281/zenodo.21881512, permanently at https://doi.org/10.5281/zenodo.21881512. Sign in and open it from this page.

Bibliographic data from OpenAlex (CC0), last checked 19 November 2026. ResearchGuru does not host this paper; once you are signed in, the links open it at the publisher or the repository that does.