cs540_shavlik_fall20..

Download Report

Transcript cs540_shavlik_fall20..

CS 540 – Introduction to AI
Fall 2016
Jude Shavlik
TA: Sam Gelman
http://pages.cs.wisc.edu/~shavlik/cs540.html
Today’s Topics
• Administrivia
–
–
–
–
–
–
–
–
The Class Home Page
Moodle Piazza?
Java, Eclipse IDE, and CS 367
Course Textbook (skim Ch 1 and 2, read Sec 18.1-18.3 & Appendices A & B)
Also skim Artificial Intelligence and Life in 2030
Late HWs, exam dates?
A little about me …
Do not email me at [email protected] (use [email protected])
•
•
•
•
•
Class at Capacity – drop soon if you decide to not take; I’ll let more in next Tues
Class Style
Some AI History and Philosophy
HW0 – Reading in a Dataset for Machine Learning (last 15 mins today)
Machine Learning (in Lecture 2)
•
Read the Algorithm section of the Wikipedia page on Random Forests (for next week) and
(soon) Pedro Domingos' paper A Few Useful Things to Know About Machine Learning
(you can access this paper for free if you are on a UW-Madison network; if you use DoIT's VPN
I believe you can also access this from a non-UW network, such as a computer in your home)
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 2
AI  Magic
• Often mathematically complex algorithms
• But lots of data (“big data”) and simple(r)
methods can work quite well!
• Counting lots of things can lead to
intelligent behavior (HW4)
• Arguably AI, especially machine learning
(ML), most important IT technology
currently – still quite exciting after being it
in for 30+ years!
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 3
Class Style
• More about engineering intelligent s/w
than modeling human cognition
• Concrete focus, to provide context for
general AI ideas
• “Hands on” – learn more by (actively)
doing than (passively) listening
• Try to write some notes during class, even
though Powerpoint of lectures available
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 4
Work Load
• About five HWs (35% of grade),
two with substantial programming
• Midterm (30%) – 90 mins, in class, Oct 20
• Final (35%) – Dec 21, 7:45am
• I will teach more than I can test;
don’t just focus on what is graded!
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 5
You Remember
(Metcalf, 1997 and others)
• 10% of what you read
• 20% of what you hear
• 30% of what you see
• 50% of what you hear and see together
• 70% of what you think and say out loud
• 90% of what you do
• ??% of what you hear, see, and write down
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 6
Programming
Knowledge Assumed
• For loops, arrays, scanners (to read in data)
• Object-oriented design (eg, trees)
• Stacks, queues, linked lists, hash tables
• Recursion (see “Programming Knowledge Assumed” slide)
• Trees and recursion (cs 367 topics)
will be a big part of HW1
• Math: partial derivatives (for neural nets);
mathematical logic and prob covered in class
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 7
HW Schedule (tentative)
HW0 – due next Tuesday, read in dataset for ML
HW1 – learn decision trees (Java)
HW2 – tune sets (Java), ensembles (Java), searching
for solutions (paper-and-pencil)
Midterm
HW3 – probabilistic reasoning (Java) and
case-based reasoning (paper-and-pencil)
HW4 – artificial neural networks and support vector
machines (paper-and-pencil)
HW5 – logical rep & reasoning (paper-and-pencil)
Final
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 8
Detailed List of Course Topics
Learning from labeled data
Experimental methodologies for choosing parameter
settings and estimating future accuracy
Decision trees and random forests
Probabilistic models,
Nearest-neighbor methods
Genetic algorithms
Neural networks
Support vector machines, kernels
Reinforcement learning (if time permits)
Learning from unlabeled data (if time permits)
K-Means
Expectation-Maximization
Searching for solutions
Heuristically finding shortest paths
Algorithms for playing games like chess
Simulated annealing
Genetic algorithms
Reasoning probabilistically
Probabilistic inference
Bayes' rule, Bayesian networks
9/5/16
Reasoning from concrete cases
Cased-based reasoning
Nearest-neighbor algorithm
Kernels
Reasoning logically
First-order predicate calculus
Representing domain knowledge using mathematical logic
Logical inference
Probabilistic logic (if time permits)
Problem-solving methods based on the biophysical world
Genetic algorithms
Simulated annealing
Neural networks
Philosophical aspects
Turing test
Searle's Chinese Room thought experiment
The coming singularity
Strong vs. weak AI
Societal impact of AI
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 9
Late HW's Policy
• HW's due on-line in Moodle @ 11:55pm
• You have 5 late days to use
over the semester
(Fri 11:55pm → Mon 11:55pm is ONE late "day")
• SAVE UP late days!
• Penalty points after late days exhausted
• Can't be more than ONE WEEK late so
solutions can be posted
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 10
Academic Misconduct
(also on course homepage)
All examinations, programming assignments, and written
homeworks must be done individually. Cheating and
plagiarism will be dealt with in accordance with
University procedures (see the Academic Misconduct
Guide for Students). Hence, for example, code for
programming assignments must not be developed in
groups, nor should code be shared. You are encouraged
to discuss with your peers, the TAs or the instructor
ideas, approaches and techniques broadly, but not at a
level of detail where specific implementation issues are
described by anyone. If you have any questions on this,
please ask the instructor before you act.
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 11
Some AI Milestones
• Computer beats leading
chess grand master (1997)
• Computer wins Jeopardy (2011)
• Speech recognition in smartphones (2011)
• Self-driving cars (2014)
• ‘Star Trek telephone’ (2015)
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 12
The “Star Trek” Telephone
Japanese
Speech Recognition
Machine Translation
Speech Generation
English
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 13
CS 540 at the Movies
(suggest better/other videos)
• AI Algo Beats Kasparov at Chess
https://www.youtube.com/watch?v=NJarxpYyoFI
• IBM Watson Wins Jeopardy
https://www.youtube.com/watch?v=WFR3lOm_xhE
• Stanford+Google Car
http://www.ted.com/talks/sebastian_thrun_google_s_driverless_car
• Microsoft SKYPE Translator
https://www.youtube.com/watch?v=mWTySUGXR2k&list=PLD7HFcN7L
XRd4kd2XgZjIbQ8TwTC32Zc9&index=3
• CS 540 Nannon© Competition
https://www.youtube.com/watch?v=b1SqrjuPrmE
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 14
DARPA Grand Challenge
(2005)
• Oshkosh Truck came in 5th
• 4th: a Louisiana insurance company!
– Story about searching for best path through
dried lake bed …
– Many fine paths, too much time spent thinking!
• What is the key difference between Chess
and Jeopardy/Car-Driving?
– ‘closed’ vs. ‘open’ world
• Can you write a progam that is smarter than you?
– You likely will in cs540
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 15
Some More Videos/Images
• Robots Falling Down at the
2015 DARPA Robotics Challenge
https://www.youtube.com/watch?v=g0TaYhjpOfo
• Google Translate
(2015 cellphone app)
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 16
In which Year will Children Born that
Year not Need to Learn How to Drive?
Recently a leading robotics researcher
said his answer is ‘2014’
• Robots too polite?
Eg, never speed, always yield
• Will existing cars be retrofitted?
• Will airplanes (especially freight) and
trucks be first? Cargo ships?
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 17
Machine Learning is
Becoming Ubiquitous
• Search (in the Google/Bing/etc sense)
• Credit-card scoring, finance in general
– Why might “hadBankruptcy” be the best feature for
deciding who gets a credit card?
• Personalization/recommendation in various forms
• Extracting ‘knowledge’ from ‘natural’ languages
(Machine Reading)
• Understanding pictures and videos, face recognition
• ML large focus of CS 540
(overlap with CS 760, grad ML class)
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 18
An AI Axiom
The easier something is for humans the
harder it is for computers
And vice versa
• A point I’ve been making for  25 years,
but maybe no longer true?
• Human-machine cooperation appealing
• AI (rapidly) replacing ‘white collar’ jobs?
(Robots have been replacing ‘blue collar’ jobs for awhile)
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 19
The Coming Singularity?
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 20
Some Interesting Quotes
• “Machine intelligence is the last invention
that humanity will ever need to make.”
http://www.ted.com/talks/nick_bostrom_what_happens_when_our_computers_get_smarter_than_we_are
• “When thinking about the future we tend to
over estimate the impacts in the near-term
and under estimate impacts in the long
term.” Roy Amara, Institute for the Future
(http://www.iftf.org/home/)
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 21
Predicted Amount of Change
Linear Thinking
9/5/16
How technology
actually advances
How we tend to
predict the future
Time into the Future
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 22
More AI Philosophy
in Final Class this Term
• Turing Test
• Searle’s Chinese Room story
• Weak and Strong AI hypotheses
• The future of AI including its societal impact
• Additional AI classes at Wisconsin
9/5/16
CS 540 - Fall 2016 (Shavlik©), Lecture 1
Lecture 1, Slide 23
HWO – Reading in an Dataset
• Due in one week (most HWs will have two weeks
between when assigned and when due)
• The Wine Dataset (original version)
9/6/15
CS 540 - Fall 2015 (Shavlik©), Lecture 2
24