To Do
- Email Glenn
- Think (hard) about evolving networks, and the representation and computation issues with regard to these measures that we’re interested in
- Write down the measures we’re interested in
- Create a list of the measures, the group of measures that we’re interested in and that make sense (in or out of our domains, but domain-specific relations would be helpful)
- Along with the measures write down the computation issues that could arise with each
- Write a list of the expanse of our domain and problems
- Write down the modes of variation (what counts as evolution)
- Characteristics of nodes change
- Characteristics of edges change
- Edges come and go
- Nodes come and go
- …
- Types of Time
- Event-driven time
- Wall-clock time
- Consider all temporal ways any measure could be computed over the evolving network
- All time
- One time point (snapshot)
- Interval
- Has a start point, end point, and temporal-width
- Changing Interval (perhaps sliding over time, sliding window)
- Contiguous intervals
- Non-contiguous intervals
- Union of intervals (contiguous and non)
- Comparison across graphs
- Variations that make comparison hard
- Comparison of intervals
- Are they in lock step?
- Do the events happen at the same time?
- Are they both using wall clock time with the same time-step?
- Are they the same length?
- Do the intervals in the two different graphs under comparison have different, independent trajectories?
- If one is a fast-event time and the other is a slow-event time, is it fair to compare them? Is it able to compare them? Is it a fair comparison?
- Get out the whole spectrum of evolving characteristics and variations that we are considering
- What can people look at for their datasets and networks that will say whether or not our work is a reasonable thing to consider computing over their data. Or if it is even possible.
- Getting out all the expanses could bring to light some good measures that will fit into these spaces
- Also for the proposal: put some bounds on things
- “We’re not doing this”
- Limit what we’re considering
- We’re not solving everyone’s network problems, only this subset of issues
Agenda
- Committee Members: Dave, Alf, Gabe, Worthy, Glenn (need to email him)
- Who would be a good person to add as a minor representative?
- New diagrams
- Thoughts?
- Does it convey the story?
- Sundry
- Finding errors in the database: people mislabeled from data entry
- Joseph should be finishing the adoptions, I still don’t have them yet
- Should we alter the database to have this new marriage events table?
- Proposal timeline
- Would like to have the final draft to committee before wedding, no later than Nov 7, 2014.
- Propose early-to-mid December
- Research questions
- Should I look into submitting viz to a CS conference, as Luther suggested? Would I need to present something different than the workshop?
Pre-Notes
Other great datasets to consider:
- Web graph, where nodes consist of companies
- Companies come and go, also buy, create, and retire websites
- Citeseer citation graph, using universities/corporations as nodes, co-authorship as edges
Other notes:
- Rewrite chord to advance to next event
- Can we avoid storing snapshots? Using snapshots??
Notes
- Possible Committee Members and Ideas
- Jeff Holt, in Statistics
- Hongning Wang (New CS Faculty): would be good to talk to, but not necessarily add to the committee (discuss with him his work with Bing)
- Dan Keenan, in Statistics
- Good contacts
- Karen Kafadar (Chair of Statistics)
- Amber Tomas, postdoc in Statistics interested in evolving networks
- Why stats would be interesting to work with
- Wall clock versus event time
- Intervals and measures over intervals
- Intervals that change and have pieces/changes of the network fall off the beginning of the interval
- Trajectories over the interval(s)
- Distributions over times, and the characteristics of these distributions
- How much change in attributes (of a node or edge) needs to happen before there is a change in identity (of that node or edge)?
- When does the node/eedge change it’s character?
- This change is then an imposed event on the graph (the change is an event)
- If we relax or tighten the identity changing function, then that affects the events in the graph
- Identity change is a function of attributes in the node (or edge)
- What are the measures we’re talking about? This NEEDS to be hammered down NOW so that I can have a chance of proposing this fall and writing a proposal document before the wedding.
- Measures over an interval
- What about letting the interval move across time?
- Incremental calculation sounds like a good candidate here
- Obviously could naively recalculate the measure, but incremental updating would be much better
- Expanding and contracting the window/interval. What would that look like?
- Proposal has two main functions
- Working committee, giving ideas and engaged in the problem
- One-sided contract: if you do what they’ve agreed on, it’s enough. But other things could also be considered enough if you continue discussions with the committee passed the proposal.
- Look toward writing a succinct statement of the thesis
- After writing some measures and the computational issues therein, start thinking about combining them into a thesis:
- This measure is interesting in this application, and you can do it in this manner.
- Can become sort of an algorithms thesis at this point.
- Do you want to move in the algorithms direction?
- Ex: Statements of time complexity or lower bounds, “you can’t do it faster than $O(-)$
- Ex: Can amortize the costs over …
- Ex: If you have $V$ nodes and $E$ edges, then you can expect …
- Start writing down the list (see above)