Efficient and Robust Bayesian Selection of Hyperparameters in Dimension Reduction for Visualization

TL;DR

This paper presents a Bayesian optimization-based auto-tuning framework for hyperparameters in dimension reduction algorithms, improving efficiency and robustness for large datasets and multiple metrics.

Contribution

It introduces a versatile, data-driven hyperparameter selection method that incorporates normalization and subsampling, applicable to visualization techniques like t-SNE and UMAP.

Findings

Effective hyperparameter tuning on large datasets

Versatile multi-objective optimization capability

Improved visualization quality metrics

Abstract

We introduce an efficient and robust auto-tuning framework for hyperparameter selection in dimension reduction (DR) algorithms, focusing on large-scale datasets and arbitrary performance metrics. By leveraging Bayesian optimization (BO) with a surrogate model, our approach enables efficient hyperparameter selection with multi-objective trade-offs and allows us to perform data-driven sensitivity analysis. By incorporating normalization and subsampling, the proposed framework demonstrates versatility and efficiency, as shown in applications to visualization techniques such as t-SNE and UMAP. We evaluate our results on various synthetic and real-world datasets using multiple quality metrics, providing a robust and efficient solution for hyperparameter selection in DR algorithms.