Friday, March 28, 2008

things any good (CS) lab should do

1. encourage almost ALL phd students to go on internships to industry labs at least once and possibly twice

2. have a really wide net for visiting talks which is broad church (include industry and social aspects)

3. involve PhDs in teaching and paper reviewing

4. take students and post docs out for dinner from time to time, and to the pub a lot

5. teach phds about processes (papers, PCs, conference management systems, grant applications) viz
http://www.cl.cam.ac.uk/~jac22/talks/jon-cfip.ppt

Friday, March 21, 2008

piety - irrational numbers and social network rganisation

pi is an interesting number - so is that fact that social groups are structured around tiers of trust, centered on each person with nearest trust group being kith and kin, and then
close friends etc...the degree of this graph is something like 3.x (e.g.
3.1415957...)
somewhere at 5 hops (3.x^5) you reach the limit of human's ability to infer intentionality

this tells us that a social networking tool (TM) should be built around making levels
and group at each power visible and should change how it supports people outside of 5 hops and groups larger than 150 or so...

i think such a tool could be called piety since that has PI in it and should be easy peasy to use

ageing (churn in friendship groups) could determine when fols are evicted from the close circle of friends into the outer darkness of cyberspace (or facebook:) and
frequency and duration of contacts (including email/IM, co-location) can be used to include people in...

coud make a nice startup:)

Thursday, March 13, 2008

references for money, and conference reviews for fun

1. so here's my idea of reference writing

I only every write 3 references - the person asking me for one as to pick one of my pre-canned references, and I will only sign if it what they pick matches my belief about them - oh, and i charge 1 euro (stable currency of choice)

2. here's my idea for conference reviewing and PC management

we have PC meetings after all the reviews are written - how about we haver the meeting just after all the papers are submitted and (modulo conflicts) the pC meet to choose who reviews what - this
a) calibrates them
b) socialises them
c) load balances (from each according to her needs, to each according to his means)

then we dont need another meeting - everything else can be done online

3. choosing PCs - easy - ask potential PC members (mostly from pool of PC members from this or related conferences in previous years) to furnish th PC chairs (and conference organising ctttee if there is one) with 2 or 3 example reviews of papers accepted in previous years - these are used to judge if the potential PC members understand the job (fairness, accuracy, etc).

4. alternative, use my reference scheme above - there are only really 9 different kind of reviews (7 if you are in CS theory) so we publish those, and authors submitting papers pick which reviews they think it, and the PC reviewer signs it if she agrees, and doesn't if he disagrees.

Tuesday, March 11, 2008

More on Mobile Social Networks

so lots of people think social networks are worth lots of money

and lots of people think mobile phones are worth lots of money

what can you do with a mobile phone in terms of social networks

1. obvious (I wrote this ages ago) is run a browser (flock) on yr smart phone and go to a social net site
2. obvious (everyone wrote this a zillion years ago) - run a location service and map people's geo-relationships to each other or to locations or both (see MIT reality mining, Intel's project doing location service using just about anything a few years back etc etc)
3. do pocket switch net and decentralised link discovery (non obvious, but we did it in pre-haggle project work about 4-5 years back) and build temporal graph in a distributed way, then use it for
i) establishing relationships (co-lo duration and frenquency above some threshold)
ii) to actualyl forward info (person to person or pub/sub, via percolation, gossip, epidemic, but filterered by relevance (e.g. interest/tags), and by trust/tribal/dunbar/closeness etc) to get efficient delivery and focus...

This is gonna be the topic for the Million People project, and is discovery (not invention) and therefore not patentable:)

clever algorithms to make it work relaibly are more interesting though...