INTERNET OF THINGS, HISTORIA OCH
TILLÄMPNINGARKary Främling, Professor i
Datavetenskap
FRÅN SAKERNAS INTERNET TILLINTELLIGENTA PRODUKTER
Transp.comp. B Transp.comp. A Destination Transp.comp. C
Set ID@URI
ID@URI
Manufacturer’s tracking agent at URI
Location updates
Manufacturer ID@URI
ID@URI
ID@URI
Virtual Counterpart (Digital Twin) år 2003Internet of Things (IoT) år 2002FRÄMLING, Kary, HOLMSTRÖM, Jan, ALA-RISKU, Timo, KÄRKKAINEN, Mikko. Product agents for handling information about physical objects . Report of Laboratory of Information Processing Science series B, TKO-B 153/03, Helsinki University of Technology, 2003. 20 p.
LITE INTELLIGENTARE PRODUKTERElectrical motor, manufactured by XXX, previous maintenance ...
Pizza containing ..., last date of use ..., heat oneminute at 600W
Bus stop XXX, next bus toyour destination arrives in three minutesHur vi gjorde IoT år 2002
SMARTA(RE) BILAR
2005
SMARTA HUSHÅLLSMASKINER
2007
House owner, inhabitants
Power monitoring
Weather station
Smart Grid
Local power production
Ventilation, refrigerators, ...
Energy prices, forecasts, control commands etc.
Data acquisition, visualisation, analysis, control, etc. (server/cloud)
SMARTA HUS 2009
SMARTARE BRUK AV FORDON
2010
SMARTARE TILLVERKNING
QualityManager
ProductionEngineer
FactoryOperator
O-MI Node
O-MI Node
Historical data for analysis
Error alarmmessage to remedy the errorimmediately
Monitor realtime data
O-MI Node
O-MI Node
Data from otherlifecycle phases
and sourcesEvent Detection
2013
SMARTA STÄDER 2016
SMART HUS TALAR MED SMART BIL I EN SMART STAD
2018
SMART BIL KOMMERTILL SMART STAD
Street temperature is…
Just drove into pothole!
ESP system activated, slippery!!
Find parking place close to …
Need to charge my EV battery, N kWh…
Parking/ EV charging placereserved…
Slippery 100 meters ahead!
Hole in road 100 meters ahead!
School children crossing in 5 minutes, change route!2020
• Varför nekades jag kredit?• Varför körde du på cyklisten, autonoma bil?• Varför får jag sådan här reklam?• Explainable AI
INTELLIGENS KRÄVER SJÄLVINSIKT
20??
• Utvecklades i Kary Främlingsdoktorsavhandling 1996
• Kan tillämpas på vilken typs ”Black Box AI” som helst
• Matematiska begrepp som har sinbakgrund i klassisk beslutsfattningsteori
• Möjligt att använda lämpligterminologi enligt användare ochsituation
• Förklaringar kan också vara visuella, via vibrationer, ljud, …
• Går också att förklara med olikanivåer av abstraktion
CONTEXTUAL IMPORTANCEAND UTILITY
Price (x1)
Room (x2)Location (x3)
Food (x4)
Hotel 1 (y=0.75)Hotel 2 (y=0.18)
Hotel 3(y=0.07)
∑
∑
∑
.
.
.
.
.
y1
y2
y3
x2
x1
x3
x4
Criteria
Black-box model
Recommendations
CIU Explainer Hotel 1 is recommended because it has a very good room with fair price.
The location of the hotel and food are ranked above average.
Hotel 2 is NOT recommended becausewhile, it has an excellent room
with an attractive price, The location of the hotel and food
are ranked very poor.
Hotel 1 versus Hotel 2
Visual and text-based explanations
PriceLocation
FoodRoom
Price Location
FoodRoom
I can explain!
Anjomshoae, Främling, Najjar. Explanations of Black-Box Model Predictions by Contextual Importance and Utility. Submitted to Explainable AI Workshop of AAMAS 2019 conference.
SMARTARE OCH BÄTTRE SAMHÄLLE?