This is a blog for students enrolled in CSCI 8980 special topic course on Big Data Algorithms in Fall 2013 at University of Minnesota to share their feedback on the course, have a forum for discussion and create a repository of datasets and relevant references. All course related information will be posted here.
Here is the course description
For decades researchers across different disciplines of computer science have envisioned the need of techniques to handle data-intensive computing. With the boom of internet and the explosion of data in every socio-economical aspect, once what was a futuristic research, has now transformed itself into a dire requirement. Big Data comes with immense opportunity, but turning this seriously high volume, high velocity, structured or unstructured, heterogeneous, often noisy and high-dimensional data into something one can use is a huge challenge. This course aims at timely dissemination of foundational algorithmic developments for big data analysis and exposing students to cutting edge research in this area. The course will involve deep theoretical analysis with the goal of developing practical algorithms with variety of applications. We will explore trade-off s among space, time and accuracy for algorithm design. The course will cover different sampling methodologies, streaming algorithms where only a small fraction of data can be stored, semi-streaming model that allows a few sequential access to disk and some parallel algorithms, specifically, the map-reduce paradigm. We will study the algorithmic and complexity aspects of these frameworks. A primary focus of this course will be on designing scalable, sub-quadratic, often near-linear or even sub-linear algorithms. We will explore property testing methodology that allows in sub-linear time to test whether a data set has certain property. We will go through the recent progress in developing fast algorithms for basic graph problems such as max-flow min-cut, matching etc., sparse transformations such as sparse fast Fourier transform, and hashing methodologies involving min-hash and locality sensitive hashing. The other topics will include dimensionality reduction techniques to handle high-dimensional data comprising of random projection method, Johnson Lindenstrauss Lemma, metric embedding and graph sparsi fiers. We will also explore crowdsourcing–the power of human assisted computing. Most of the algorithms that we will study in this course will crucially use randomization and will give an answer that is a good approximation of the optimal solution.
Prerequisites Students should have basic knowledge of algorithms: running time analysis, graphs algorithms, and must be familiar with discrete probability. Undergraduates are welcome to attend if they satisfy the requirements.
Syllabus There will be no required text book for this course, instead we will use assorted materials from the web. A tentative list of topics is as follows:
- Streaming: Sampling and Sketching
- Dimensionality Reduction
-
External Memory and Semi-streaming Algorithms
- Map-Reduce Framework and Extensions
-
Near Linear Time Algorithm Design
- Property Testing
-
Metric Embedding
- Sparse Transformation
- Crowdsourcing
The class meets every Thursday 6:30 pm–9:00 pm CT at MechE 221.
Office hours: Friday 1pm to 5pm at 6-198 KHKH by Appointment.
See pdf flyer of the course with contact information Algorithmic Techniques for Big Data Analysis
Grading Policy
Scribe 20% + Class and Blog Participation 20% + Project 60% (Survey 30% + Presentation and Write up 30%)
For each Thursday class,scribe is due by the following Monday. Late submission will be penalized.
* Oct 3rd : submit names of 5 survey papers
* Oct 17th: project proposal due – Max 2 pages + references, must be in latex, single column, article 11 pt.
*Nov 14th : a one page write-up describing progress on project due
* Dec 5th: project final write-up due
No late submission will be accepted unless there is compelling reason.
Relevant Other Courses Taught in Other Universities
- Sub-linear Algorithms by Piotr Indyk, Ronitt Rubinfeld at MIT http://stellar.mit.edu/S/course/6/sp13/6.893/
- Algorithms for Massive Data Sets by Piotr Indyk at MIT http://people.csail.mit.edu/indyk/MASS/index.html
- Mining Big Data by Anand Rajaraman and Jeffrey D. Ullman at Stanford University http://infolab.stanford.edu/~ullman/mining/2009/index.html
- Dealing with Massive Data by Sergei Vassilvitskii from Google Research at Columbia University http://www.cs.columbia.edu/~coms699812/
- Data Stream Algorithms by Andrew Mcgregor at UMASShttp://people.cs.umass.edu/~mcgregor/courses/CS711S12/index.html
- Algorithms for Big Data Management by Ashwin Machanavajjhala at Duke https://www.cs.duke.edu/courses/spring13/compsci590.2/
- Data Stream Algorithms by Amit Chakrabarti at Dartmouth http://www.cs.dartmouth.edu/~ac/Teach/CS49-Fall11/
Lectures:
Latex template for writing scribe scribe
Lecture notes will be emailed to the students unedited and will be posted here after I have time to edit them at least once.
- 09/05 Overview Slides Various models and techniques, counting distinct items in different models, no scribe.
- 09/05 Introduction to data streams, finding frequent items deterministically, lower bound for deterministic computation of distinct elements, Markov inequality, Chebyshev bound, The Chernoff bound. Scribe by Vivek Mishra. Lecture 1
- 09/12 Universal family of hash functions, Analysis of two algorithms for counting distinct items Slides
- 09/12 Count-Min sketch and applications Slides from Andrew Mcgregor’s Data Streaming Course, Project discussion Project Discussion
- 09/19 Frequency moment estimation, Johnson-Lindenstrauss Lemma, p-stable distribution. l_p norm via max-stability “High frequency moments via max-stability”. Slides Scribe
- 09/19 Approximate Near Neighbor Search, Locality Sensitive Hashing Slides Scribe
- 09/26 Locality Sensitive Hashing continued, Min-wise Independent Hashing , note: http://cseweb.ucsd.edu/~dasgupta/254-embeddings/lawrence.pdf
- 09/26 Sampling: Reservoir, AMS, Priority sampling Slides
- 10/03 Clustering: Small space clustering: K center, K median; K-means++ Slides, k-means++
- 10/03 Semi-streaming: Graph connectivity, Spanners, Sparsifiers Slides by Andrew Mcgregor
- 10/10 Graph Sparsifiers Continued
- 10/10 Graph Sketch http://people.cs.umass.edu/~mcgregor/711S12/lec-2-2.pdf
- 10/17 Making Dynamic Programmings Fast: Amnesic Approximation. A tour through Data Quality. Data Quality Slides
- 10/17 Introduction to Map Reduce, MST, Matching, Dense Subgraph Computation Map-Reduce Intro Slides
- 10/24 Counting Triangles in Massive Graphs
Student presentation: Radius and Clustering on Map Reduce
Student presentation: Data Deduplication, Load balancing in Map reduce - 10/31 Random Walks on Graphs
Student presentation: Distributed Random Walks - 10/31 Recommender System: Low Rank Matrix Completion
- 11/07 Student Presentation: Survey on Recommender System, Real time recommendation
- 11/07 Student Presentation: Managing Uncertain Data: Ranking with uncertainty
- 11/14 Crowdsourcing: Ranking and Clustering
- 11/14, 11/21 Student Presentation: Topics on Social Streams
Tag Recommendation, Story Detection in Twitter - 12/5, Property Testing & Sublinear Algorithms: Testing Distance Between Distributions, Counting Connected Components, MST
- Project Presentation
Background on Randomized Algorithms:
- Randomized Algorithms by Motwani and Raghavan
- Probability and Computing by Mitzenmacher and Upfal
- The Probabilistic Method by Alon and Spencer
Relevant Reading
- Sketch Techniques for Approximate Query Processing by Graham Cormode
- The space complexity of approximating the frequency moments, Noga Alon, Yossi Matias, Marios Szegedy
- Stable Distributions, Pseudorandom Generators, Embeddings and Data Stream Computation, Piotr Indyk
- An elementary proof of a theorem of Johnson and Lindenstrauss, Sanjoy Dasgupta, Anupam Gupta
- Approximate nearest neighbors: towards removing the curse of dimensionality by Piotr Indyk and Rajeev Motawani
- Near-Optimal Hashing Algorithms for Approximate Nearest Neighbor in High Dimensions by Alex Andoni and Piotr Indyk.
- Locality-sensitive hashing scheme based on p-stable distributions, Mayur Datar, Nicole Immorlica, Piotr Indyk, Vahab S. Mirrokni
- Priority sampling for estimation of arbitrary subset sums by Nick Duffield, Carsten Lund, Mikkel Thorup
- Streaming Algorithms from Precision Sampling, Alexandr Andoni, Robert Karuthgamer, Krzysztof Onak
- Tight Results for Clustering and Summarizing Data Streams, Sudipto Guha
- Better streaming algorithms for clustering problems, Moses Charikar, Liadan O’Callaghan, Rina Panigrahy
- k-means++: The Advantage of Careful Seeding, Sergei Vassilvitskii and David Arthur
- On Graph Problems in Semi-Streaming Model, Joan Feigenbaum, Sampatha Kannan, Andrew Mcgregor, Siddharth Suri.
- Graph Distances in the Data-Stream Model, Joan Feigenbaum, Sampatha Kannan, Andrew Mcgregor, Siddharth Suri, Jian Zhang.
- Graph sparsification in the semi-streaming model, Kook Jin Ahn, Sudipto Guha
- Bahman Bahmani, Ravi Kumar, Sergei Vassilvitskii. Densest Subgraph in Streaming and MapReduce. In VLDB 2012
- Silvio Lattanzi, Benjamin Moseley, Siddharth Suri, Sergei Vassilvitskii. Filtering: A Method for Solving Graph Problems in MapReduce. In SPAA 2011
- Howard Karloff, Siddarth Suri, Sergei Vassilvitskii. A Model of Computation for MapReduce. In SODA 2010 (Austin, Texas) [ pdf]
Algorithmic Techniques for Big Data Analysis 