From Robustness to Efficiency Yin Tat Lee
Description: From Robustness to Efficiency Yin Tat Lee University of Washington Theme Theme How designing robust algorithms lead to fast algorithms? Example 1: Gradient Descent Not a precise result because the convergence depends on the function
Related Topics
Download Presentation
"From Robustness to Efficiency Yin Tat Lee" is the property of its rightful owner. Permission is granted to download and print the materials on this website for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Presentation Transcript
slide1. From Robustness to Efficiency Yin Tat Lee
University of Washington<br>
slide2. Theme<br>
slide3. Theme How designing robust algorithms lead to fast algorithms?<br>
slide4. Example 1: Gradient Descent * Not a precise result because the convergence depends on the function class.<br>
slide5. Example 2: Ball Walk (Mangoubi-Vishnoi 19) Lazy Update
When the algorithm is robust, we can delay some calculations. (Lovász 90)<br>
slide6. “Newton method”<br>
slide7. Linear Programs Linear Systems Consequence of Robust Interior Point Method<br>
slide8. Robust Interior Point Method Newton method can made to be robust to input.
Cohen-Lee-Song 19
v.d.Brand 20<br>
slide9. Formula for Newton Method<br>
slide10. Robust Interior Point Method<br>
slide11. Update when necessary and batch update when possible
Cohen-Lee-Song 19
v.d.Brand 20<br>
slide13. # of coordinate changes Ignored normalization in norms and many log terms<br>
slide14. Batch Update<br>
slide15. Other Results<br>
slide16. Other Applications / Related Work<br>
slide17. Open Problem: LP in input sparsity time<br>
University of Washington<br>
slide2. Theme<br>
slide3. Theme How designing robust algorithms lead to fast algorithms?<br>
slide4. Example 1: Gradient Descent * Not a precise result because the convergence depends on the function class.<br>
slide5. Example 2: Ball Walk (Mangoubi-Vishnoi 19) Lazy Update
When the algorithm is robust, we can delay some calculations. (Lovász 90)<br>
slide6. “Newton method”<br>
slide7. Linear Programs Linear Systems Consequence of Robust Interior Point Method<br>
slide8. Robust Interior Point Method Newton method can made to be robust to input.
Cohen-Lee-Song 19
v.d.Brand 20<br>
slide9. Formula for Newton Method<br>
slide10. Robust Interior Point Method<br>
slide11. Update when necessary and batch update when possible
Cohen-Lee-Song 19
v.d.Brand 20<br>
slide13. # of coordinate changes Ignored normalization in norms and many log terms<br>
slide14. Batch Update<br>
slide15. Other Results<br>
slide16. Other Applications / Related Work<br>
slide17. Open Problem: LP in input sparsity time<br>