CS395T: Sublinear Algorithms (Fall 2014)

Logistics: Tue/Thu 2:00 - 3:30
GDC 4.304
Unique Number: 53295
Course web page: http://www.cs.utexas.edu/~ecprice/courses/sublinear/
Professor: Eric Price
Email: ecprice@cs.utexas.edu
Office: GDC 4.510
Office Hours: Tuesday 3:30-5, Wednesday 3-4
Useful References: Similar courses include Sublinear Algorithms (at MIT), Algorithms for Big Data (at Harvard), and Sublinear Algorithms for Big Datasets (at the University of Buenos Aires).
Content: This graduate course will study algorithms that can process very large data sets. In particular, we will consider algorithms for:
  • Data streams, where you don't have enough space to store all the data being generated.
  • Property testing, where you don't have enough time to look at all the data.
  • Compressed sensing, where you don't have enough measurement capacity to observe all the data.
Problem Sets:
  1. Problem Set 1. Due September 23.
  2. Problem Set 2. Due October 9.
  3. Problem Set 3. Due October 28.
  4. Problem Set 4. Due November 11.
  5. Problem Set 5. Due November 25.
Lectures:
  1. Thursday, August 28. Course overview; uniformity testing; beginning of distinct elements. [Lecture notes (pdf) (tex)]
  2. Tuesday, September 2. Concentration of measure; Markov, Chebyshev, subgaussians, subexponentials, Johnson-Lindenstrauss. [Lecture notes (pdf) (tex)]
  3. Thursday, September 4. Continue distinct elements; streaming turnstile model; AMS-sketch. [Lecture notes (pdf) (tex)]
  4. Tuesday, September 9. Count-Min, Count-Sketch, Fourier analysis of Count-Sketch. [Lecture notes (pdf) (tex)]
  5. Thursday, September 11. More Count-Sketch; sparse recovery with sublinear time. [Lecture notes (pdf) (tex)]
  6. Tuesday, September 16. L0 sampling. Graph sketching preliminaries. [Lecture notes (pdf) (tex)]
  7. Thursday, September 18. Graph sketching.
  8. Tuesday, September 23. Coresets. [Lecture notes (pdf) (tex)]
  9. Thursday, September 25. Fp estimation.
  10. Tuesday, September 30. Covering and packing numbers, RIP.[Lecture notes (pdf) (tex)]
  11. Thursday, October 2. Compressed Sensing, Iterative Hard Thresholding.[Lecture notes (pdf) (tex)]
  12. Tuesday, October 7. Model-Based Compressed Sensing, L1 minimization.[Lecture notes (pdf) (tex)]
  13. Thursday, October 9. l2/l1 upper bound. deterministic l2/l2 lower bound. randomized l1/l1 lower bound.[Lecture notes (pdf) (tex)]
  14. Tuesday, October 14. Project ideas. RIP-1. SSMP.[Lecture notes (pdf) (tex)]
  15. Thursday, October 16. Adaptivity, group testing, entropy.[Lecture notes (pdf) (tex)]
  16. Tuesday, October 21. Communication complexity and information cost.[Lecture notes (pdf) (tex)]
  17. Thursday, October 23. Adaptive sparse recovery, introduction to Fourier transform.[Lecture notes (pdf) (tex)]
  18. Tuesday, October 28. Getting familiar with Fourier transforms.
  19. Thursday, October 30. Nonequispaced Fourier transforms.
  20. Tuesday, November 4. Sparse Fourier transforms.
  21. Thursday, November 6. RIP of subsampled Fourier.[slides]
  22. Tuesday, November 11. Oblivious Subspace Embeddings.[Lecture notes (pdf) (tex)]
The tentative outline for the rest of the course is as follows:
  • Sparse JL
  • Bloom filters; invertible bloom lookup tables
  • Property testing: uniformity testing
  • More property testing
  • Other streaming models: random order, distributional
Prerequisites: Mathematical maturity and comfort with undergraduate algorithms and basic probability. Ideally also familiarity with linear algebra.
Grading: 40%: Homework
30%: Final project
20%: Scribing lectures
10%: Participation
Scribing: In each class, one student will be assigned to take notes. These notes should be written up in a standard LaTeX format before the next class.
Homework
policy:		 There will be a homework assignment roughly every two weeks.

Collaboration policy: You are encouraged to collaborate on homework. However, you must write up your own solutions. You should also state the names of those you collaborated with on the first page of your submission.
Final project: In lieu of a final exam, students will perform final projects. These may be done individually or in groups of 2-3. An ideal final project would perform a piece of original research in a topic related to the course. Failing that, one may perform a literature survey covering several research papers in the field.

Students will present their results to the class during the last week of classes. The final paper will be due on the scheduled final exam day.

Students with
Disabilites: 		Any student with a documented disability (physical or cognitive) who requires academic accommodations should contact the Services for Students with Disabilities area of the Office of the Dean of Students at 471-6259 (voice) or 471-4641 (TTY for users who are deaf or hard of hearing) as soon as possible to request an official letter outlining authorized accommodations.