Nonlinear Contrastive Dimension Estimation
Background: Contrastive dimension reduction seeks to extract signals unique to a foreground group relative to a background group, a process that frequently involves selecting an appropriate reduced-dimensional space. Determining this dimension parameter is currently often reliant on ad-hoc heuristics or computationally expensive tuning methods.
Question / Future Work: Developing a principled, data-driven approach to estimate the contrastive dimension in nonlinear settings remains an open challenge. Current methods struggle to adapt linear-based dimension estimation to scenarios where data reside on complex, curved manifolds, hindering the reproducibility and automated selection of this critical hyperparameter.
Why It Matters: Selecting the reduced dimension is a primary bottleneck for practitioners; addressing this would improve the usability, reliability, and scientific reproducibility of CDR across various fields.
Evidence: As a result, even in linear CDR, choosing d in a principled and reproducible way remains an open problem. While these issues are already substantial in the linear setting, they become even more pronounced when the data are believed to lie on curved manifolds… A reliable notion of contrastive dimension in the nonlinear setting could directly inform the choice of the reduced dimension parameter d.
Metadata & Links
- created_at
- 2026-04-22T05:44:43Z