2006-12-11

• What:
• Who: Miller Puckette
• http://crca.ucsd.edu/~msp/

• wrote Max program for computer music.
• says computer music is fustrating because no real jumps in development
• Ended up loosing control of Max, so he started over from scratch.
• interested in computers as musical instruments.
• this lecture feels like "soft science."
• Speaker is having trouble communicating. He seems to have interesting ideas, but I can't understand them. He's not very clear.

2006-12-04

• What: HCI
• Who: ?? from georgia tech

• synergy of hci, se, and vision
• Classroom 2000 - classroom with whiteboard, video recorders. After class, slides with notes were pasted on the web. Selecting annotations forwarded the video to that point.
• blah blah blah

2006-11-13

• What: Bayesean filtering
• Who: Michael Jordan

2006-10-23

• What: how to find an advisor
• Who: panel
  • Lavelle
  • Carey - HCI
  • Darko - automated software testing
  • Campell - Social computing
• how do you go about finding an advisor?
  • email a professor (better than going to the person's door.)
  • talk to other students
  • some people post their requirements
  • looking for someone who is really motivated, looking for someone who makes the contact.

2006-10-16

• What: Duality in projective geometry
• Who: Dana S Scott, emeritus professor from CMU
  • dana.scott@cs.cmu.edu
• http://mathworld.wolfram.com/ProjectiveGeometry.html
• Undecidable theories?
• consider the space of n-degree polynomials. This forms a space, differentiaiton is a projection. How can we utilize the duality theorem of projective geometry in order to say things about polynomial differentiation.

• in essence, he's using geometry to prove things about symbolic differentiation
• he's defining a "geometrical space" where each "point" is a function
• and then using geometry theorems on that
• and in this space, differentiation is a geometrical transformation/projection that takes points from one thing to another.
• so he can use X theories, blah blah blah

2006-10-09

• subvirt: Implementing malware with virtual machines.
  • Sam King
• Financial incentives for taking over computers.
• previously, attackers and defenders fought at various privledge levels in the operating system
• now, people are installing their own virtualization levels below the OS
  • bypasses all security measures availiable to the public

2006-10-02

Automated unit testing

• code testing
  • generation
  • execution
  • verification
• at microsoft, 75% of time spent testing.
• generate large data set, automate testing for structually complex data.
• idea - certain data structures have constraints that have to be met at all time
  • generator generates lots of inputs
  • code transforms inputs to outputs
  • oracle verifies the structure
• challenge - generate valid tructured inputs.
  • describe validity properties
  • efficiently generate test data (input should be sparse)
• Explored an academic specificiation language
  • all these speciality language were too hard to learn
• Korat
  • predicates written in target language
  • need to generate inputs from predicates (hard!)
• bounded exhaustive generation
  • put a limit on the input space. Need values and nubmer of objects
  • genreate all valid imputs within a given bound, test
  • avoid isomorphic inputs (permutations of nodes?)
  • naive soln:
    • generate entire input space
    • run program on each input
    • not feasible for sparse spaces
  • key idea
    • capture all variable references, use that information to prune search of input space.
  • use Java Modeling language to specify pre/post conditions, invariants {{{ //@ invariant }}}
• used at microsoft, plenty of other places

2006-09-18

• Who: Kevin C. Chang
• What:

Talk is all about data mining the deep web. It presumes data has a defined structure that you can specify ahead of time.

Won't adapt to new structures.

2006-09-11

• Who: Dmitri Willims
• What: Entrance, voice, loyalty: Social dynamics in online gamine communities.

http://blogs.parc.com/playon/ - blog about sociology of gaming.

2006-08-28

• Who: Indranil Gupta (Indy for short)
• What: "Tipping points" for districured systems

(I think this is the new guy who works beside Kale's lab. Apparently he's hot stuff.)

interesting, use gossip as a model for distributing information.

tipping point - that moment when an idea, trend, or cidial behaio crosses a threshold, and spreads like wildfire.

program on a per node basis, generate emergent behavior.

very much like swarm intellegence.

contagon theory - you will follow a trend if X% of your neighbors are following a trend.

2006-08-23

• Who: Cheng Xiang Zhai
• What: UCAIR - user centered adaptive information retrieval

Looks like the search agent that I want! I need to ask what resources they have for keeping track of searching sessions.

Idea - collect information with search agents on the users computers. Makes personalization much easier than centralizes services like google.

rank queries with documents by comparing word probablilty distribution with query word probability distribution.

move the evaluation of queries to the client side, where the user information is.

use snippets for analysis, as the snippets are what attracted the user in the first place.

they have a toolbar! Neat! I have to get access to it.

• Key ideas
  • user modeling
  • serach context modeling
  • information integration
  • optimixing ever resopons in interactive IR
• Results
  • implicit user modeling can improve search accuracy
  • text mining can provide user specific search results