https ps://ww ://www.cs.ub w.cs.ubc. c.ca ca/s /stud tudent - - PowerPoint PPT Presentation

https ps ww cs ub w cs ubc c ca ca s stud tudent ents s
SMART_READER_LITE
LIVE PREVIEW

https ps://ww ://www.cs.ub w.cs.ubc. c.ca ca/s /stud tudent - - PowerPoint PPT Presentation

Departme partment nt of Computer puter Science nce Undergr dergrad aduat uate e Events ts for Sept 10-14 14 Mor ore e details ls @ https ps://ww ://www.cs.ub w.cs.ubc. c.ca ca/s /stud tudent ents/ s/und under ergra grad/l


slide-1
SLIDE 1

Departme partment nt of Computer puter Science nce Undergr dergrad aduat uate e Events ts for Sept 10-14 14 Mor

  • re

e details ls @

https ps://ww ://www.cs.ub w.cs.ubc. c.ca ca/s /stud tudent ents/ s/und under ergra grad/l d/life ife/up /upco comin ming-ev event ents

TELUS US Info Sessi sion

  • n

Date: e:

  • Mon. Sept. 10

Time: e: 5:30 – 7:30 pm Locati cation

  • n: Wesb

sbroo

  • ok 100

100 Deloitt

  • itte

e Info Se Sess ssion

  • n

Date: e: Tues.

  • s. Sept. 11

Time: e: 6:00 – 8:00 pm Locati cation

  • n: Henr

nry y Angus us Room m 098 Capgem pgemini ini Info Sessio sion Date:

  • Fri. Sept. 14

Time: e: 2:00 – 5:00 pm Loca cation

  • n: Downtown

ntown Va Vancou couver ver (RSVP SVP req’d by Sept. 12) Tri-Me Ment ntori

  • ring

ng Student ent Orientatio ntation Date: e: Tues.

  • s. Sept. 11

Time: e: 5 5:15 – 6:30 pm Locati cation

  • n: DMP 110

Resu sume e Writing ing Work rkshop shop (for r non- coop

  • ps)

s) Date: e: Thur urs.

  • s. Sept. 13

Time: e: 12:30 0 – 1:45 pm Locati cation

  • n: DMP 101

101 Wome men n in Games es Panel Date: e: Wed. . Se Sept 12 Time: e: 5:30 – 9:00 pm Locati cation

  • n: EA Bu

Burna naby by Studios

  • s
slide-2
SLIDE 2

CPSC 322, Lecture 3 Slide 2

AI I App pplica catio tions ns

Computer ter Sc Science ce cpsc3 c322 22, , Lectur ture e 3

Sept, t, 10, 2012

slide-3
SLIDE 3

CPSC 322, Lecture 2 Slide 3

Modules dules we'l 'll l cover er in th this course: se: R&Rsys sys

En Enviro ronm nmen ent Pr Problem em

Query Planning Deterministic Stochastic Search Arc Consistency Search Search Value Iteration

  • Var. Elimination

Constraint Satisfaction Logics STRIPS Belief Nets Vars + Constraints Decision Nets Markov Processes

  • Var. Elimination

Stati atic Sequenti ntial al Representation Reasoning Technique

slide-4
SLIDE 4

CPSC 322, Lecture 2 Slide 4

Modules dules we'l 'll l cover er in th this course: se: R&Rsys sys

En Enviro ronm nmen ent Pr Problem em

Query Planning Deterministic Stochastic Search Arc Consistency Search Search Value Iteration

  • Var. Elimination

Constraint Satisfaction Logics STRIPS Belief Nets Vars + Constraints Decision Nets Markov Processes

  • Var. Elimination

Stati atic Sequenti ntial al

slide-5
SLIDE 5

CPSC 322, Lecture 3 Slide 5

(Ad Adve versarial) rsarial) Se Search ch: : Checkers ckers

Gam ame e pl play ayin ing was one of the first

tasks undertaken in AI Ar Arthur ur Sa Samuel at IBM wrote programs to play checkers (1950s)

  • initially, they played at a strong

amateur level

  • however, they used some (simple)

machine learning techniques, and soon outperformed Samuel

Source: IBM Research

Chinook’s program was declared the Man- Machine World Champion in checkers in 1994! …and complete etely ly solve ved by a program in 2007!

slide-6
SLIDE 6

CPSC 322, Lecture 3 Slide 6

(Ad Adve versarial) rsarial) Se Search ch: : Chess ss

In 1996 and 1997, Gary Kasparov, the world chess grandmaster played two tournaments against Deep Blue, a program written by researchers at IBM

Source: IBM Research

slide-7
SLIDE 7

CPSC 322, Lecture 3 Slide 7

(Ad Adve versarial) rsarial) Se Search ch: : Chess ss

Deep Blue’s Results in the first tournament:

  • won 1 game, lost 3 and tied 1

first time a reigning world champion lost to a computer

Source: CNN

slide-8
SLIDE 8

CPSC 322, Lecture 3 Slide 8

(Ad Adve versarial) rsarial) Se Search ch: : Chess ss

Deep Blue’s Results in the second tournament:

  • second tournament: won 3 games, lost 2, tied 1
  • 30 CPUs + 480 chess processors
  • Searched 126.000.000 nodes per sec
  • Generated 30 billion positions per move reaching

depth 14 routinely

slide-9
SLIDE 9

Sample mple A* applications lications

  • An

An Ef Efficie cient nt A* A* Se Search Al Algorith ithm Fo For St Statistical stical Machine Translation. 2001

  • Th

The General aliz ized ed A* A* Ar Architec tectur

  • ture. Journal of

Artificial Intelligence Research (2007)

  • Machine Vision … Here we consider a new

compositional model for finding salient curves.

  • Fa

Factor tored d A* A*searc rch h for models s over sequences nces and trees International Conference on AI. 2003…. It starts saying… The primary challenge when using A*

search is to find heuristic functions that simultaneously are admissible, close to actual completion costs, and efficient to calculate… applied to NLP and BioInformatics

CPSC 322, Lecture 9 Slide 9

slide-10
SLIDE 10

CPSC 322, Lecture 2 Slide 10

Modules dules we'l 'll l cover er in th this course: se: R&Rsys sys

En Enviro ronm nmen ent Pr Problem em

Query Planning Deterministic Stochastic Search Arc Consistency Search Search Value Iteration

  • Var. Elimination

Constraint Satisfaction Logics STRIPS Belief Nets Vars + Constraints Decision Nets Markov Processes

  • Var. Elimination

Stati atic Sequenti ntial al

slide-11
SLIDE 11

CPSC 322, Lecture 3 Slide 11

CSP SPs: s: Crossword ssword Pu Puzzl zzles es

Source: Michael Littman

slide-12
SLIDE 12

CPSC 322, Lecture 3 Slide 12

CSP SPs: s: Radio io link k fr frequency uency ass ssignment ignment

Source: INRIA

Assigning frequencies to a set of radio links defined between pairs of sites in order to avoid d interfe rfere renc nces es. Constraints on frequency depend on positio tion of the links ks and on physi sica cal l enviro ronme ment nt . Sample Constraint network

slide-13
SLIDE 13

13

Ex Example le: : SL SLS f S for RNA A secon

  • nda

dary ry structu cture re design

RNA strand made up of four bases: cytosine (C), guanine (G), adenine (A), and uracil (U) 2D/3D structure RNA strand folds into is important for its function Predicting structure for a strand is “easy”: O(n3) But what if we want a strand that folds into a certain structure?

RNA strand

GUCCCAUAGGAUGUCCCAUAGGA

Secondary structure Easy Hard

On of the Best algorithm to date: Local search algorithm RNA-SSD developed at UBC [Andronescu, Fejes, Hutter, Condon, and Hoos, Journal of Molecular Biology, 2004]

CPSC 322, Lecture 1

slide-14
SLIDE 14

Constraint nstraint optimi timizatio zation n problems blems

Optimization under side constraints (similar to CSP) E.g. mixed integer programming (software: IBM CPLEX)

  • Linear program: max cTx such that Ax ≤ b
  • Mixed integer program: additional constraints, xi  Z (integers)
  • NP-hard, widely used in operations research and in industry

Transportation/Logistics: Supply chain Production planning SNCF, United Airlines management and optimization: UPS, United States software: Airbus, Dell, Porsche, Postal Service, … Oracle, Thyssen Krupp, SAP,… Toyota, Nissan, ...

14 CPSC 322, Lecture 1

slide-15
SLIDE 15

CPSC 322, Lecture 2 Slide 15

Modules dules we'l 'll l cover er in th this course: se: R&Rsys sys

En Enviro ronm nmen ent Pr Problem em

Query Planning Deterministic Stochastic Search Arc Consistency Search Search Value Iteration

  • Var. Elimination

Constraint Satisfaction Logics STRIPS Belief Nets Vars + Constraints Decision Nets Markov Processes

  • Var. Elimination

Stati atic Sequenti ntial al

slide-16
SLIDE 16

CSP SP/lo /logic: gic: fo formal al ve verification ification

16

Hardware verification Software verification

(e.g., IBM) (small to medium programs) Most progress in the last 10 years based on: Encodings into propositional satisfiability (SAT)

CPSC 322, Lecture 1

slide-17
SLIDE 17

CPSC 322, Lecture 3 Slide 17

Logic: gic: Cyc ycSecure Secure

“scans s a computer ter netwo work rk to build a f formal mal represe sent ntati ation

  • n of

the network, based on Cyc’s pre-existing ontology of networking, security, and computing concepts: This formal representation also allows users to interact directly with the model of the network, allowing testing of proposed changes.”

Excerpted from: Shepard et al., 2005

  • Kn

Knowl wledge dge Repres esen entat tatio ion

  • Se

Semantic tic Web !

slide-18
SLIDE 18

CPSC 322, Lecture 2 Slide 18

Modules dules we'l 'll l cover er in th this course: se: R&Rsys sys

En Enviro ronm nmen ent Pr Problem em

Query Planning Deterministic Stochastic Search Arc Consistency Search Search Value Iteration

  • Var. Elimination

Constraint Satisfaction Logics STRIPS Belief Nets Vars + Constraints Decision Nets Markov Processes

  • Var. Elimination

Stati atic Sequenti ntial al

slide-19
SLIDE 19

CPSC 322, Lecture 3 Slide 19

Pl Planning anning & Sc & Scheduling: eduling: Logistics istics

Dynamic Analysis and Replanning Tool (Cross & Walker)

  • logistics planning and scheduling for military transport
  • used in the 1991 Gulf War by the US
  • problems had 50,000 entities (e.g., vehicles); different

starting points and destinations

Source: DARPA

Same techniques can be used for non-military applications: e.g., Emergency Evacuation

slide-20
SLIDE 20

CPSC 322, Lecture 3 Slide 20

Pl Planning: anning: Sp Spacecraft cecraft Control trol

NASA: Deep Space One spacecraft

  • perated autonomously for two days in May, 1999:
  • determined its precise position using stars and

asteriods

despite a malfunctioning ultraviolet detector

  • planned the necessary course adjustment
  • fired the ion propulsion system to make this adjustment

Source: NASA

For another space application see the Spike system for the Hubble telescope

slide-21
SLIDE 21

Source: cs221 stanford

Slide 21 CPSC 322, Lecture 1

slide-22
SLIDE 22

CPSC 322, Lecture 2 Slide 22

Modules dules we'l 'll l cover er in th this course: se: R&Rsys sys

En Enviro ronm nmen ent Pr Problem em

Query Planning Deterministic Stochastic Search Arc Consistency Search Search Value Iteration

  • Var. Elimination

Constraint Satisfaction Logics STRIPS Belief Nets Vars + Constraints Decision Nets Markov Processes

  • Var. Elimination

Stati atic Sequenti ntial al

slide-23
SLIDE 23

CPSC 322, Lecture 3 Slide 23

Reasoning asoning under der Unce certai rtainty: nty: Diagnosis gnosis

Source: Onisko et al., 99

Bayes Net: to diagnose liver diseases

slide-24
SLIDE 24

CPSC 322, Lecture 3 Slide 24

Source: Mike Cora, UBC

Reasoning asoning Under der Unce certai rtainty nty

Texture classification using Support Vector Machines

  • foliage, building, sky, water
slide-25
SLIDE 25

Reasoning asoning Under der Unce certai rtainty nty

E.g. motion tracking: track a hand and estimate activity:

  • drawing, erasing/shading, other

Source: Kevin Murphy, UBC

Slide 25 CPSC 322, Lecture 1

slide-26
SLIDE 26

Com

  • mpu

puter ter Vis isio ion n (no not t ju just t fo for rob

  • bot
  • ts!)

!)

Jing, , Baluja, Ro Rowl wley,

, Goo

  • ogl

gle: e: Fi Find ndin ing g Can anon

  • nic

ical al Im Imag ages es

Slide 26 CPSC 322, Lecture 1

Source: cs221 stanford

slide-27
SLIDE 27

Compare mpare low-level level fe features tures

Slide 27 CPSC 322, Lecture 1

Source: cs221 stanford

slide-28
SLIDE 28

In Induced duced Gr Graph ph

Slide 28 CPSC 322, Lecture 1

Source: cs221 stanford

slide-29
SLIDE 29

AI I - Mach chine ine Learning rning @google

  • gle
  • Spam/Porn Detection
  • Which ad to place given a query
  • Train Speech to search on mobile
  • Machine Translation
  • …..

CPSC 322, Lecture 1 Slide 29

  • Highly Parallelizable EM + Map Reduce (simple

code to write)

  • Stochastic Gradient Descent
slide-30
SLIDE 30

Watso son : analyzes natural language questions and content well enough and fast enough to compete and win against champion players at Jeopardy!

CPSC 322, Lecture 1 Slide 30

Source: IBM

“This Drug has been shown to relieve the symptoms

  • f ADD with relatively few side effects."
  • 1000s

0s of algorit rithms hms and KBs KBs, ,

  • 3

3 secs secs

slide-31
SLIDE 31

St Statistical tistical Mach chine ine Tr Translat slation ion

SEHR GEEHRTER GAST! KUNST, KULTUR UND KOMFORT IM HERZEN BERLIN. DEAR GUESTS, ART, CULTURE AND LUXURY IN THE HEART OF BERLIN. DIE ÖRTLICHE NETZSPANNUNG BETRÄGT 220/240 VOLT BEI 50 HERTZ. THE LOCAL VOLTAGE IS 220/240 VOLTS 50 HZ. DE EN

Source: cs221 Stanford

Slide 31 CPSC 322, Lecture 1

slide-32
SLIDE 32

信 letter trust letters believe signal a letter believe that letter of confidence 说 自己 themselves said that say they said he say that said they themselves saying that he would say that said that she had saying that he has 仍 然 是 continues to be are still the main would still be continued to be remains one of remains one continues to be the still is remains an area still viewed by are always one of 是 总理 Prime Minister the Prime Minister is the Prime Minister 他 he He

  • ther

his him

  • ther

that he he was him to he is he has

  • f his

他 信 Thaksin Thaksin Chinnawat and Joint Communique Dr Thaksin Joint Communique , Mr Thaksin in his letter his letter

  • thers

他 信 也 Thaksin also 总理 , 拒绝 …… 辞 职 . resign . leaving their service .

  • f leaving their service .

resigned as counsel .

他 信 也 说 自己 仍 然 是 总理 , 拒 绝 辞 职 .

Source: cs221 stanford

Slide 32 CPSC 322, Lecture 1

slide-33
SLIDE 33

Zi Zite: te: a a personalize sonalized d magazine azine

… that gets smarter as you use it

CPSC 322, Lecture 1 Slide 33

slide-34
SLIDE 34

CPSC 322, Lecture 2 Slide 34

Modules dules we'l 'll l cover er in th this course: se: R&Rsys sys

En Enviro ronm nmen ent Pr Problem em

Query Planning Deterministic Stochastic Search Arc Consistency Search Search Value Iteration

  • Var. Elimination

Constraint Satisfaction Logics STRIPS Belief Nets Vars + Constraints Decision Nets Markov Processes

  • Var. Elimination

Stati atic Sequenti ntial al

slide-35
SLIDE 35

CPSC 322, Lecture 3 Slide 35

Decision Network in Finance for venture capital decision

Source: R.E. Neapolitan, 2007

slide-36
SLIDE 36

CPSC 322, Lecture 3 Slide 36

Pl Planning anning Under er Unce certainty rtainty

Source: Jesse Hoey UofT 2007

Learning and Using POMD

MDP

models of Patient-Caregiver Interactions During Activities

  • f Daily Living

Goal: l: Help Older adults living with

cognitive disabilities (such as Alzheimer's) when they:

  • forget

et the proper r sequence ce of tasks that need to be completed

  • they lose track

k of the steps that they have already completed.

slide-37
SLIDE 37

CPSC 322, Lecture 3 Slide 37

Pl Planning anning Under er Unce certainty rtainty

Helicopter control: MDP, reinforcement learning St States: es: all possible positions, orientations, velocities and angular velocities Final solution involves Deterministic search ch!

Source: Andrew Ng 2004

slide-38
SLIDE 38

CPSC 322, Lecture 1 38

Military litary applications: lications: eth thica cal l iss ssues ues

  • Robot soldiers
  • Existing: robot dog carrying

heavy materials for soldiers in the field

  • The technology is there
  • Unmanned airplanes
  • Missile tracking
  • Surveillance
slide-39
SLIDE 39

Dec ecision ision Th Theo eory: : Dec ecis ision ion Su Supp ppor

  • rt

t Sy Systems stems

E.g., Computational Sustainability New interdisciplinary field, AI is a key component

  • Models and methods for decision making concerning the management

and allocation of resources

  • to solve most challenging problems related to sustainability

Often constraint optimization problems. E.g.

  • Energy: when are where to produce green energy most economically?
  • Which parcels of land to purchase to protect endangered species?
  • Urban planning: how to use budget for best development in 30 years?

39

Source: http://www.computational-sustainability.org/

CPSC 322, Lecture 1

slide-40
SLIDE 40

CPSC 322, Lecture 3 Slide 40

Dim imensi ensions

  • ns of
  • f Rep

epres esen entationa tational l Com

  • mpl

plex exity ity in in CPSC32 322

We'v 've already dy discu cuss ssed ed:

  • Deterministic versus stochastic domains
  • Static versus sequential domains

So Some other r importan tant t dimensi sion

  • ns

s of complex exity: ity:

  • Explicit state or propositions or relations
  • Flat or hierarchical
  • Knowledge given versus knowledge learned from

experience

  • Goals versus complex preferences
  • Si

Sing ngle le-ag agen ent t vs. . mu mult lti-agent agent

slide-41
SLIDE 41

CPSC 322, Lecture 3 Slide 41

Multiagent ltiagent Sy Syst stems: ems: Po Poke ker

“In full 10-player games Poki is better er than a typical cal low-limi imit casino ino play ayer er and wins consistently; however, not as good

  • d as most expert

perts s New programs being developed for the 2-player game are quite a bit better, and we believe they will very soon surpass all human players”

Source: The University of Alberta GAMES Group

Sear arch ch Space ce: 1.2 quintillion nodes

slide-42
SLIDE 42

CPSC 322, Lecture 3 Slide 42

Multiagent ltiagent Sy Syst stems: ems: Robot

  • t So

Soccer ccer

Source: RoboCup web site

Extre remely ely complex mplex

  • Stochastic
  • Sequence of actions
  • Multiagent

robotic soccer competition was proposed by LCI (UBC) in 1992 (which became Robocup in 1997).

slide-43
SLIDE 43

CPSC 322, Lecture 3 Slide 43

TO TO DO O fo for Next t class ss

  • Search: Start reading (Chpt 3 – sec 3.1 – 3.3)
  • If your stude

dent nt ID is below come and talk k to me

  • 2201909

9095, 5, 13301114 114