Related Experiment Videos
MAPLE: Self-Supervised Learning-Enhanced Nonlinear Dimensionality Reduction for Visual Analysis
Summary
We introduce MAPLE, a new nonlinear dimensionality reduction technique that improves upon UMAP by using self-supervised learning for better manifold modeling. MAPLE enhances data visualization for complex datasets like biological and image data.
Area of Science:
- Machine Learning
- Data Science
- Bioinformatics
Background:
- Nonlinear dimensionality reduction is crucial for visualizing high-dimensional data.
- Existing methods like UMAP may struggle with complex manifold structures and intra-cluster variance.
- Improved manifold modeling is needed for clearer data representation.
Purpose of the Study:
- To introduce MAPLE, a novel nonlinear dimensionality reduction method.
- To enhance manifold modeling using self-supervised learning and maximum manifold capacity representations (MMCRs).
- To demonstrate MAPLE's effectiveness on high-dimensional biological and image data.
Main Methods:
- Developed MAPLE, a self-supervised learning approach for dimensionality reduction.
- Utilized maximum manifold capacity representations (MMCRs) to untangle complex manifolds.
- Employed qualitative and quantitative evaluations to compare MAPLE with UMAP.
Main Results:
- MAPLE demonstrates improved manifold modeling compared to UMAP.
- The method effectively handles high-dimensional data with substantial intra-cluster variance and curved structures.
- MAPLE achieves clearer visual cluster separations and finer subcluster resolution.
Conclusions:
- MAPLE offers enhanced dimensionality reduction by improving manifold modeling.
- The technique provides superior visualization of complex datasets, particularly biological and image data.
- MAPLE maintains computational tractability while delivering improved performance over UMAP.
