Daphne Koller - PowerPoint Presentation

Slide1

Daphne KollerStanford University

Learning

to improve our livesSlide2

Input

Computers Can Learn?

Computers can learn to predictComputers can learn to act

Output

Program

Parameters

Learned to get desired input/output mappingSlide3

Many, many, many applicationsSpeech recognition

Fraud detectionIntrusion detection into computer systemsImage searchActivity recognition in video surveillanceAutonomous driving (DARPA Grand Challenge)Early epidemic detectionCancer subtype classificationUncovering basic biological mechanisms….Slide4

Example: Spam FilteringSpam email:Comprises 85-95% of email trafficCost US organizations > $13 billion in 2007

Spammers are constantly adapting hand-constructed systems bound to failSlide5

pharmacy

unique offerdoctor! (x 2)…

Input

Learning to Detect Spam

Features

Output

Program

Parameters

5

7

0.5

1

…

“

spamness

”

=

18.3

Learned to optimize prediction quality

Increase parameters for words appearing in spam email

Decrease parameters for words appearing in good

email

Can learn in advance

Online learning adapts to changing trends

And to personalize to a user’s preferences

Collaborative learning allows learning from other people’s dataSlide6

Harder: Machine TranslationInput can’t be viewed as a “bag” of wordsOutput is not a simple decision (spam / not spam) but a complex sentence

Machine translation using human-constructed translation rules floundered for decadesThe spirit is willing but the flesh is weak.The vodka is good but the meat is rotten.

English to Russian and backSlide7

Thanks to: Mehran

SahamiHarder: Machine TranslationML-based machine translation

systemsUse matched text in two languages to learn matching between words or phrasestext in target language to learn what “good” text is likeSlide8

PerceptionImpossible using hand-coded rulesExample: Automated handwriting recognition

Deployed at all 250+ Postal Distribution Centers25 billion+ letters processed annually> 92% automated processingHundreds of millions of $ saved each year

Thanks to:

Venu

GovindarajuSlide9

LearnedProgram

Multi-Sensor Integration: Traffic

Trained on historical data

Learn to predict current &

future

road speed, including on

unmeasured

roads

Dynamic route optimization

Multiple

views

on traffic

Incident reports

Weather

Thanks to: Eric Horvitz

I95 corridor experiment: accurate to



5 MPH in 85% of cases

Fielded in 72 citiesSlide10

Controlling Complex Systems

Thanks to: Andrew NgSlide11

Controlling Complex SystemsLearning by emulating a human (apprenticeship)… and by adapting to experience

Adjust parameters to reward good behaviorThanks to: Andrew NgSlide12

Future: Smart Power GridKey problem: Get (clean) energy from where it’s produced to where it’s needed on limited grid

Solution: LearningPerception: predicting current and future demandsControl: Make robust and efficient routing decisionsSlide13

Thanks to: Eric Horvitz

Medical DiagnosisImprove quality of diagnosis:Computer diagnosis systems outperform most doctorsAllow triage by less-experienced peopleSlide14

Medical InterventionPatient-specific automatic detection of epilepsy seizures from EEG for real-time intervention

Thanks to: John Guttag

Seizure Onset

patients

Response latency (sec)

Generic approach

Per-patient learned modelSlide15

Medical InterventionPatient-specific automatic detection of epilepsy seizures from EEG for real-time intervention

Reduce frequency of medical errorsLearn “standard of care” and detect anomaliesReduce enormous cost: financial and human lifeHome-based systems for tracking of chronic patients for early prediction of complicationsReduce pain, suffering, and cost of hospitalizationSlide16

Scientific DiscoveryNew technologies revolutionize biologyHigh-throughput sequencing

Gene expressionProtein-protein interactionsProteomicsCellular microscopy….But how do these help understand & cure disease?Slide17

Humans differ in 0.1% of their DNAThese differences determine who we are, what diseases we’ll get, and which cures will work for usWhich differences matter?

Our Genes Determine Who We Are

Diabetes patients

Healthy

individuals

…ACTCGGTGGGCATAAATTCGGCCCGGTCAGATTCCATCCAGTTTGTTCCATGG…

…ACTCGGTGGGCATAAATTCGGCCCGGTCAGATTCCATCCAGTTTGTACCATGG…

…ACTCGGTGGGCATAAATTCGGCCCGGTCAGATTCCATCCAGTTTGTACCATGG…

: :

…ACTCGGTGGGCATAAATTCGGCCCGGTCAGATTCCATCCAGTTTGTACCATGG…

…ACTCGGTGGGCATAAATTCTGCCCGGTCAGATTCCATCCAGTTTGTTCCATGG…

…ACTCGGTAGGCATAAATTCGGCCCGGTCAGATTCCATACAGTTTGTACCATGG……ACTCGGTGGGCATAAATTCGGCCCGGTCAGATTCCATACAGTTTGTTCCATGG……ACTCGGTAGGCATAAATTCGGCCCGGTCAGATTCCATACAGTTTGTACCATGG… : …ACTCGGTGGGCATAAATTCTGCCCGGTCAGATTCCATCCAGTTTGTACCATGG…

…ACTCGGTGGGCATAAATTCTGCCCGGTCAGATTCCATACAGTTTGTTCCATGG…Slide18

Only 5% of DNA appears to play functional roleTo understand which genetic changes matter, we need to find the functional pieces, such as genesTrain model using known genesLearn what DNA sequences characterize them

Where Are the Genes?Thanks to: Michael Brent

Machine learning critical to gene findingSlide19

Future: Smart HealthcareEvidence-based medicine: Learn what works… at personalized level: What works

for meLearn mapping from individual genotype and other factors to disease risk and drug suitability Slide20

Machine Learning = Computing on Steroids

Data

Challenging

Application

Machine

Learning

ML core technology for prediction and decision

Makes possible applications where other methods simply don’t work

Perception

Personalization

Dynamic adaptation

Can improve almost any application

A little bit of learning goes a long way