MEDALLION: Manifold Embedding Distillation and Attributions for Low-dimensional Latent InterpretatiON

Abstract

Dimensionality reduction is widely used to visualize high-dimensional data and guide scientific discovery, yet interpreting which original features contribute to low-dimensional representations remains challenging, particularly for black-box methods such as t-SNE and UMAP. We introduce \textbf{MEDALLION} (Manifold Embedding Distillation and Attributions for Low-dimensional Latent InterpretatiON), a novel framework for interpreting low-dimensional embeddings through knowledge distillation and neural feature attribution. MEDALLION distills any given embedding with a neural encoder which we explain via a broad range of existing neural feature attribution methods. Through experiments on MNIST and applications in single-cell transcriptomics, we show MEDALLION provides informative feature attributions that cast additional insights into the underlying embedding and compare its performance with existing interpretability approaches.

Gromov-Wasserstein Distillation for Inductive Multi-View Embedding
Rafael Pereira Eufrazio, Eduardo Fernandes Montesuma, Charles Casimiro Cavalcante
arXiv30 Sep 2026PreprintarXiv licenceFull textStatisticsarXiv
Obstructions in Symplectic Embedding Problems for Four-Dimensional Convex Toric Domains
Theodore Coyne
arXiv5 Oct 2026PreprintCC BY 4.0Full textGeometry & topologyarXiv
Spectral embedding through weak* limit of finite-dimensional approximations
Fabrice Nonez
arXiv24 Oct 2026PreprintCC BY 4.0Full textAnalysis & PDEs Logic & foundationsarXiv
The Metagame of Interpretability and Meta-Attributions
Hubert Baniecki, Przemyslaw Biecek, Fabian Fumagalli
arXiv6 Oct 2026PreprintCC BY 4.0Full textStatisticsarXiv
Diffusion Distillation for Efficient Weather Ensembles
Yiming Yang, Valentin Brekke, James Briant +1
arXiv27 Aug 2026PreprintCC BY 4.0Full textStatisticsarXiv

Cite

BibTeX RIS

BibTeX

@misc{jung2026medallion,
  title = {{MEDALLION: Manifold Embedding Distillation and Attributions for Low-dimensional Latent InterpretatiON}},
  author = {Vicky Jung and Claire M. He and Irene Chang and Genevera I. Allen},
  howpublished = {arXiv},
  year = {2026},
  month = nov,
  doi = {10.48550/arxiv.2611.08767},
  eprint = {2611.08767},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.08767},
  note = {CC BY 4.0},
}

RIS

TY  - UNPB
TI  - MEDALLION: Manifold Embedding Distillation and Attributions for Low-dimensional Latent InterpretatiON
AU  - Vicky Jung
AU  - Claire M. He
AU  - Irene Chang
AU  - Genevera I. Allen
PY  - 2026
DA  - 2026/11/05
PB  - arXiv
DO  - 10.48550/arxiv.2611.08767
UR  - https://arxiv.org/abs/2611.08767
AB  - Dimensionality reduction is widely used to visualize high-dimensional data and guide scientific discovery, yet interpreting which original features contribute to low-dimensional representations remains challenging, particularly for black-box methods such as t-SNE and UMAP. We introduce \textbf{MEDALLION} (Manifold Embedding Distillation and Attributions for Low-dimensional Latent InterpretatiON), a novel framework for interpreting low-dimensional embeddings through knowledge distillation and neural feature attribution. MEDALLION distills any given embedding with a neural encoder which we explain via a broad range of existing neural feature attribution methods. Through experiments on MNIST and applications in single-cell transcriptomics, we show MEDALLION provides informative feature attributions that cast additional insights into the underlying embedding and compare its performance with existing interpretability approaches.
ER  -

Abstract

Related

Cite