Approximated and User Steerable tSNE for Progressive Visual Analytics
Download as PPTX, PDF1 like682 views
This document presents an approach called Approximate and User Steerable t-Distributed Stochastic Neighbor Embedding (A-tSNE) for progressive visual analytics of large and streaming datasets. A-tSNE speeds up t-SNE by approximating similarity computations during layout using k-nearest neighbors. It allows interactive steering and refinement of the layout. The approach is evaluated on two case studies, speeding up t-SNE over 100x on gene expression data and enabling interactive layout of streaming activity sensor data.
![8
Approximated
K-Nearest-Neighborhood [1]
Precision: 50%
[1] Fast Approximate Nearest Neighbors with Automatic Algorithm Configuration - Muja et al. - 2009
K-Nearest-Neighborhood
Approximated similarities computation](https://image.slidesharecdn.com/a-tsnevis-161028131632/85/Approximated-and-User-Steerable-tSNE-for-Progressive-Visual-Analytics-8-320.jpg)
![18[1] Hierarchical Stochastic Neighbor Embedding - Pezzotti et al. - 2016
Conclusions
• Approximation in Progressive
Visual Analytics
• Approximated-tSNE
• Data manipulation
• Refinement
• Scalability issues of the gradient
descent
• Hierarchical SNE [1]](https://image.slidesharecdn.com/a-tsnevis-161028131632/85/Approximated-and-User-Steerable-tSNE-for-Progressive-Visual-Analytics-18-320.jpg)
Approximated and User Steerable tSNE for Progressive Visual Analytics
- 1. Approximate and User Steerable tSNE for Progressive Visual Analytics Nicola Pezzotti, Boudewijn P.F. Lelieveldt, Laurens van der Maaten, Thomas Höllt, Elmar Eisemann, Anna Vilanova
- 3. 3 Non-linear dimensionality-reduction algorithm • Preserves small neighborhoods • Reveals global structures Visualizing data using t-SNE - Van der Maaten & Hinton - 2008 t-Distributed Stochastic Neighbor Embedding
- 4. 4 tSNE Similarities Computation Similarities Computation Gradient descent minimization Similarities tSNE as a Black Box
- 5. 5 PVA - tSNE Similarities Computation Similarities Computation Gradient descent minimization Similarities Progressive Visual Analytics: User-Driven Visual Exploration of In-Progress Analytics - Stolper et al. - 2014 Opening the Black Box: Strategies for Increased User Involvement in Existing Algorithm Implementations - Muhlbacher et al. - 2014 Progressive Analytics: A Computation Paradigm for Exploratory Data Analysis - Fekete & Primet - 2016 Visualization Compute partial results tSNE Progressive Visual Analytics (PVA)
- 6. Approximated Computations in PVA
- 7. 7 Approximated - tSNE Similarities Computation Similarities Visualization Compute partial results Approximated Similarities PVA - tSNE Approximated tSNE
- 8. 8 Approximated K-Nearest-Neighborhood [1] Precision: 50% [1] Fast Approximate Nearest Neighbors with Automatic Algorithm Configuration - Muja et al. - 2009 K-Nearest-Neighborhood Approximated similarities computation
- 9. 9 Approximated - tSNE Similarities Computation Similarities Visualization Compute partial results Approximated Similarities Approx. Refinement Exact Refinement Approximated tSNE
- 10. 10 tSNE Time: 3191.8 s A-tSNE Precision: 35% Time: 30.1 s Speed up: 100x Precision 35% ?
- 12. 12 • Density-based visualization • Supports brushing & linking • Approximation is visualized and removed if requested • 3 Strategies • Local minima avoidance Steerability & Approximation visualization A-tSNE Precision: 5% Preprocessing: 12 s
- 13. Case Study I : Gene Expression in the Mouse Brain
- 14. 14 Case Study I : Gene expression Sagittal Axial 3D Volume Coronal 61164 data points (Voxels) 4345 dimensions (Gene expression)
- 15. 15 Case Study I : Gene expression A-tSNE 50 seconds – tSNE 3 hours and 50 minutes Speed up: 250x
- 16. Case Study II : High-dimensional data streams
- 17. 17 Case Study II : High-dimensional data streams Chest - Ankle - Wrist 52 Dimensions every 100 ms Image courtesy of www.activ8all.com
- 18. 18[1] Hierarchical Stochastic Neighbor Embedding - Pezzotti et al. - 2016 Conclusions • Approximation in Progressive Visual Analytics • Approximated-tSNE • Data manipulation • Refinement • Scalability issues of the gradient descent • Hierarchical SNE [1]
- 19. 19 Thank you for your attention! A-tSNE Precision: 35% tSNE A-tSNE Precision: 5% Similarities computation time: 12 sSimilarities computation time: 29 s Precomp. 3195 s Speed 4x 29 s 12 s
- 20. 20
- 21. 21
- 22. 22
- 23. 23
- 24. 24
- 25. 25
- 26. 26