A Visual Interaction Framework for Dimensionality Reduction Based Data Exploration
summary
Proposes a concrete interaction layer for dimensionality reduction that lets users change inputs and see outputs (forward projection), or push points in the 2-D embedding and synthesize inputs that achieve them (backward projection). Delivers two visual tools—prolines that precompute and show per-attribute projection paths, and feasibility maps that sample the output grid to show constraint-satisfying regions. Implements out-of-sample extrapolation for PCA and encoder/decoder nets (autoencoders) and demonstrates on an 8-D OECD set and 784-D MNIST digits.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 4:28Forward Projection
- 9:24Auto Encoder
- 12:08Backward Projection
- 15:45Visibility Maps
- 16:51Discussion Points
related talks
37mA Bluffer's Guide to Dimension Reduction
38mManifold Learning and Dimensionality Reduction for Data Visualization
42mInteractive Data Analysis: Visualization and Beyond
34mDimensionally reducing data squeezing out the good stuff
34mVisualising Multi Dimensional Data
01h 00mData, visualization, and designing AI
17mAvoiding drill-down fallacies with VisPilot: assisted exploration of data subsets
51mArt of Effective Visualization of Multi-dimensional Data