1. Introduction to Machine Learning Lecture 2 Albert Orriols i
Puig [email protected] i l @ ll ld Artificial Intelligence
Machine Learning Enginyeria i Arquitectura La Salle gy q
Universitat Ramon Llull
2. Recap of Lecture 1 Knowledge Kno ledge Search representation
We have seen several search techniques: Blind search, heuristic
search, adversary search GAs We have seen several ways of
representing our knowledge Logic-based representation, rule-based
representation g p p We have discussed reasoning mechanisms to deal
with uncertainty, incompleteness and inconsistency y p y We set the
basis. But, the most interesting is still missing Machine learning
M hi l i Slide 2 Artificial Intelligence Machine Learning
3. Todays Agenda Whats Machine Learning Why Machine Learning?
Where is ML Headed and Which Are our Goals? Slide 3 Artificial
Intelligence Machine Learning
4. Whats Machine Learning Build computer programs that
automatically improve p pg yp with experience Can you be more
precise? (Mitchell 1997) (Mitchell, Learning = Improving with
experience at some task Improve over task T I tk With respect to a
performance measure P Based on experience E B d i E.g.: Learn to
play checkers T: Play checkers P: % of games won in world
tournament E: opportunity to play against self Slide 4 Artificial
Intelligence Machine Learning
5. What Does this Involve? Represent the knowledge p g
Logic-based representation Rule-based representation Rl b d t ti
Frame-based representation Search toward better solutions Blind
search , but not really efficient! Non-systematic techniques: G
GAs, etc. Slide 5 Artificial Intelligence Machine Learning
6. Why Machine Learning? Several factors affected the
increasing appeal of ML From the machines point of view: Recent
progress in algorithms and theory Computational power is available
From the industry point of view: Growing flood of online data GB
hours of data: Remote sensors, telescopes scanning the skies,
scientific simulations Budding industry Machine learning may help
scientists, businessmen, and engineers Classify and segment data y
g Formulate hypotheses Slide 6 Artificial Intelligence Machine
Learning
7. Why Machine Learning? There are three special niches for ML:
p Data mining: extract information from historical data to help dec
s o decision making ag Medical records Extract knowledge to help
doctors Software applications that are too complex to build a hard-
wired solution for Autonomous driving g Speech recognition Self
customizing programs Recommender systems (RS) New generation RS
Slide 7 Artificial Intelligence Machine Learning
8. Whats Data Mining in a Picture 1 J. Han, M. Kamber. J Han M
Kamber Data Mining Concepts and Mining. Techniques. Morgan
Kaufmann, 2006(Second Edition) Slide 8 Artificial Intelligence
Machine Learning
9. Do You Have a Definition for DM 1 Many definitions of data
mining. A specially interesting y g p y g one is provided by Duda,
Hart, & Stork (2002) Data mining is the process of extracting
interesting useful and interesting, useful, novel information from
data Many other definitions, but for sure, data mining is not Look
up an entry in a data base Query a web search engine How this
relates to ML? ML provides methods to dig these data Slide 9
Artificial Intelligence Machine Learning
10. Example of DM 1 Ge Given 9714 patient records, each one
describing pregnancy and birth Each patient record consists of 215
features Learn to predict Classes of future patients at risk for
Emergency Cesarean Section Slide 10 Artificial Intelligence Machine
Learning
11. Example of DM 1 O e of t e u es ea ed One o the rules
learned: Slide 11 Artificial Intelligence Machine Learning
12. Example 2 of DM 1 Slide 12 Artificial Intelligence Machine
Learning
13. Example 3 of DM 1 Slide 13 Artificial Intelligence Machine
Learning
14. Example 4 of DM 1 Slide 14 Artificial Intelligence Machine
Learning
15. Other Examples of DM 1 Slide 15 Artificial Intelligence
Machine Learning
16. 2 Problems Too Difficult to Program by Hand Autonomous Land
Vehicle in a Neural Network (ALVINN) ( ) drives 70 mph on highways
Perception system which learns to control the NAVLAB vehicles by
watching a person drive Slide 16 Artificial Intelligence Machine
Learning
17. Self-Customizing Software 3 Originally at www.wisewire.com
System that delivered a unique blend of AI with collaborative and
content-based filtering Purchased by Lycos, Inc in 1998 Integrated
in Lycos products Documents search for and find interested people.
No longer available at www.wisewire.com Visit the f ll i Vi it th
following webpage for b f more information:
http://www.cse.iitb.ac.in/dbms/Data http://www cse iitb ac
in/dbms/Data /Papers-Other/Web/wisewire.html Slide 17 Artificial
Intelligence Machine Learning
18. Where is All this Headed? Today: y First-generation systems
are evolving toward competent systems that ca tackle so e important
p ob e s efficiently and scalably a can ac e some po a problems e c
e t y a d sca ab y Give me some prove of that Ask Google Yahoo
Docomo Labs Google, Yahoo, Tomorrow T Semantic networks integrated
in DM systems Can you image face book mining? DM in many decision
processes: marketing, industry, science DM as individual
recommender systems Slide 18 Artificial Intelligence Machine
Learning
19. But Slow Down! Where are we? We are still beginning! Whats
thi Wh t this course about? b t? Starting in ML, understanding the
problems that we can solve now and the f d h future problems bl
This course is not a typical ML course in which we will go through
different paradigms Engineers solve problems, so this course tries
to follow this idea by y describing important challenges presenting
one or several of the most influential techniques to address this
challenge Slide 19 Artificial Intelligence Machine Learning
20. Next Class Characteristics Desired for ML Methods Summary
of the Paradigms that We Wont Won t Study Summary of the Problems
that We Will Study Slide 20 Artificial Intelligence Machine
Learning
21. Introduction to Machine Learning Lecture 2 Albert Orriols i
Puig [email protected] i l @ ll ld Artificial Intelligence
Machine Learning Enginyeria i Arquitectura La Salle gy q
Universitat Ramon Llull