CS 595N: Computer Science Faculty Research Seminar (Winter 2015)

Organizer: Huijia (Rachel) Lin, rachel.lin(at)cs(dot)ucsb(dot)edu

Time and location: Friday 1:00-2:00pm, Harold Frank Hall 1132

Seminar webpage: http://www.cs.ucsb.edu/~rachel.lin/courses/15w595N/

Seminar Description

The computer science department continues its tradition of having faculty members introducing their exciting works to all students interested in research, especially to new PhD and master students. The purpose of this seminar is multi-fold: Bringing faculty research closer to students, fostering dialogue between faculty and students outside the boundary of research labs, extending the horizon of students at all stage of their research paths.

Set-up: This seminar takes place weekly on Fridays. Every week, two faculty members speak for 30 minutes each (1:00-1:30pm and 1:30-2:00pm) about their research.

Special PIZZA Time: Two times in the quarter, on Jan. 23rd and Feb.13th, we will have special pizza time before the seminar (12:30-1:00pm) to treat early bird attendants with pizza as lunch.

Grading: Grade is Pass or Fail, depending entirely on attendance and punctuality.

Talk Schedule

There might be changes during the quarter. Please check the schedule periodically.

WeekDateTalk 1 (1:00pm to 1:30pm) Talk 2 (1:30pm to 2:00pm)
1 2015-01-09
  • Speaker: Ambuj Singh
  • Talk: Detecting Network Processes
  • Abstract: Below
2 2015-01-16
3 2015-01-23 Special Pizza Time
  • We offer free pizza for early-bird attandents
    between 12:30pm to 1:00pm
4 2014-01-30
5 2014-02-06
3 2015-02-13 Special Pizza Time
  • We offer free pizza for early-bird attandents
    between 12:30pm to 1:00pm
7 2014-02-20
8 2014-02-27
9 2014-03-06
10 2014-03-13


Abstracts

Week 1, Jan 9th

Ambuj Singh
Detecting Network Processes

In order to understand complex networks, we need to characterize the dynamic processes occurring within them. Examples of such processes include network congestion in the Internet or a transportation network, or the spread of malware through an online social network. The detection of such dynamic processes requires a model of underlying behavior using which inferences about significance or anomalous behavior can be made. Detecting anomalies in networks is a well-understood problem when restricted to only the graph structure (e.g., communities, structural holes), but there has been limited work on networks with node/edge attributes. When node attributes are allowed to change over time, the smooth evolution of network substructures can be used to detect significant network processes that grow, shrink, and merge over time. I will discuss an approach that compares the value at a node/edge with a background distribution, and uses a positive score to indicate significance. I will discuss methods for detecting highest scoring subgraphs (fixed substructures, varying time intervals), and for detecting smoothly varying substructures. Finally, I will discuss methods that detect significant network substructures over two classes of networks in order to explain their differences. Examples will be drawn from information networks and brain networks to illustrate the methods.