DAVID KARGER. Checkered Past Core Algorithms –graph algorithms, randomization, combinatorial...

Post on 21-Dec-2015

230 views 0 download

Tags:

Transcript of DAVID KARGER. Checkered Past Core Algorithms –graph algorithms, randomization, combinatorial...

DAVID KARGER

Checkered Past

• Core Algorithms– graph algorithms, randomization, combinatorial optimization– min-cuts, max-flows, shortest paths, minimum spanning tree, TSP– Still do some work

• Applied Algorithms– Lots of collaborations– Compilers– Web caching Akamai Technologies– Peer to peer systems– Networking: Denial of Service, SPAM blocking, Censorship evasion– Coding theory: Turbo Codes, Network Coding– Machine Learning, graphical models– Natural Language Processing– Computational Biology– Can’t resist a good problem

One Big Question

How do we make it easier for regular people

to deal with information?

Capture

Organize

RetrieveQuery

Share

Annotate

Publish

Many Answers

• Use any applicable domain– HCI– Information retrieval– Databases– Machine Learning– Social networks– Crowdsourcing– Semantic Web

• Full-cycle research

• Study Users (ethnography

)

• Build Systems

• Deploy• Recruit

Users

Handling information scraps that don’t fit anywhere

Listit

Needs

1. Too much effort to write it down/put it in“If it takes three clicks to get it down, it’s easier to e-mail [to myself].”

2. No tool for it / it doesn’t fit“I wanted to assign dates to notes, but [it] would only allow dates on

tasks.”“Where else would I keep my { guitar tabs / poetry }?”

3. Too hard to organize“It’s too much work to decide which section it should go in –sometimes

things don’t fit in just one place. It’s hard to decide what to do.”

4. Visibility and availability“If it’s not in my face, I’ll forget about it.”

“I need it with me at all times, so I have no choice.”

Listit

• Minimal Tool– Firefox plugin– One-click entry/access– No organization– Text-search retrieval

• Deployed in 2009– 19000 users– 2000 study subjects– 120,000 notes

Try it! http://listit.csail.mit.edu/

Automatically handling incoming information streams

Atomate

music listened to

running

sleep

desktop

activity

physical locations

event

s documents

message

s

travelsfriends/

enemies

Design

• Parse data streams from the web

• Build a structured model of user state

• User writes rules in Controlled Natural Language

• Refers to items and properties in data model

Friendsourced Content Sharing

Feedme

Suggestions

Choose Send

Results

• People sent more• Right recipients recommended• People liked what they got

Try it! http://feedme.csail.mit.edu/

Collaborative Lecture-Note Annotation

NB

Deployment

• Used in 15 classes at MIT, Harvard• Students see/answer each others’ questions while reading• Faculty learn what’s confusing• Close study of 6.055 Fall 2010

– 100 students created 20,000 notes– Initially, hated being forced to use the tool– By end, were praising how it increased their ability to learn– Faculty found huge value in understanding what students were thinking

Try it!http://nb.csail.mit.edu/

Easy Data Publishing

Exhibit

Regular User

Pro Site

sortsort

filterfiltersearchsearch

templatetemplate

Pro Site

Exhibit

• Pro web sites require databases and programming• Exhibit lets people author data and visualization in plain

html documents• Deployed 2005• Several 100 sites• Including newspapers and other professionals

Datapress

• Wordpress plugin• Upload or link to data

– Spreadsheet, JSON….• Then WYSYWIG your

visualization– Using usual Wordpress

blog post editor

Botanist

Music Lover/Entrepeneur

Pro Blogger

More Info

http://haystack.csail.mit.edu/

Download/try most of the tools