Skip to content
DraconiticResearch
Search Sources API
201,818 studies31 sourcesUpdated 54 min ago
PreprintCC BY-SA 4.0arXiv

On the Optimisation Landscape of Overparameterised Weighted MMD

  • Tharrshann Jayan Logarajah
  • Jeremias Knoblauch
  • Louis Sharrock
Open at sourcePDFDOICite

Abstract

We study the optimisation landscape of the maximum mean discrepancy (MMD) over weighted atomic measures. For the one-dimensional Gaussian kernel, we show that, for any compactly supported target distribution, sufficiently many particles ensure that every local minimum has arbitrarily small MMD. Moreover, any stationary point whose MMD remains bounded away from zero is unstable under an unweighted gradient flow over the particle locations. These results hold both when the weights are constrained to the simplex, and when they are unconstrained but optimised at each step. When the target is itself finitely supported, we further strengthen this result: beyond an explicit particle threshold, there are no spurious local minima, and every spurious stationary point is unstable. To our knowledge, these are the first finite-particle overparameterisation guarantees of their kind for weighted MMD optimisation.

Related

  1. Une méthode d'optimisation stochastique pour évaluer des minima à ε près
    Jean-Sébastien Le Brizaut
    Bulletin des Sciences MathématiquesPeer-reviewedLicence not statedFull text2 citedGeneral mathematicsHAL
  2. A Spike-Driven Neuromorphic Framework for Asynchronous Population-Based Optimisation
    Jorge Cruz-Duarte, El-Ghazali Talbi
    IEEE Transactions on Evolutionary ComputationPeer-reviewedLicence not statedOptimisation & controlHAL
  3. Méthodes d'optimisation pour l'approche de problèmes aux limites non linéaires mixtes elliptiques hyperboliques
    Jean-Sébastien Le Brizaut
    Bulletin des Sciences MathématiquesPeer-reviewedLicence not statedFull text2 citedGeneral mathematicsHAL
  4. Optimisation of distributed storage in existing district heating networks to solve hydraulic bottlenecks
    Anne-Geneviève Lemelle, Erwan Bescher, Nicolas Lamaison +3
    Energy ReportsPreprintLicence not stated1 citedOptimisation & controlHAL
  5. A MINLP optimisation method for the integration of storage in district heating networks
    Anne-Geneviève Lemelle, Nicolas Lamaison, Nicolas Vasset +2
    Energy Conversion and Management: XPreprintLicence not statedOptimisation & controlHAL

Cite

BibTeXRIS
BibTeX
@misc{logarajah2026optimisation,
  title = {{On the Optimisation Landscape of Overparameterised Weighted MMD}},
  author = {Tharrshann Jayan Logarajah and Jeremias Knoblauch and Louis Sharrock},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.19861},
  eprint = {2610.19861},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.19861},
  note = {CC BY-SA 4.0},
}
RIS
TY  - UNPB
TI  - On the Optimisation Landscape of Overparameterised Weighted MMD
AU  - Tharrshann Jayan Logarajah
AU  - Jeremias Knoblauch
AU  - Louis Sharrock
PY  - 2026
DA  - 2026/10/14
PB  - arXiv
DO  - 10.48550/arxiv.2610.19861
UR  - https://arxiv.org/abs/2610.19861
AB  - We study the optimisation landscape of the maximum mean discrepancy (MMD) over weighted atomic measures. For the one-dimensional Gaussian kernel, we show that, for any compactly supported target distribution, sufficiently many particles ensure that every local minimum has arbitrarily small MMD. Moreover, any stationary point whose MMD remains bounded away from zero is unstable under an unweighted gradient flow over the particle locations. These results hold both when the weights are constrained to the simplex, and when they are unconstrained but optimised at each step. When the target is itself finitely supported, we further strengthen this result: beyond an explicit particle threshold, there are no spurious local minima, and every spurious stationary point is unstable. To our knowledge, these are the first finite-particle overparameterisation guarantees of their kind for weighted MMD optimisation.
ER  -