Machine Learning

Definition: The ability of a machine to improve its performance based on previous results.

Components to be learned Direct-mapping from conditions on the current state to actions Means to infer properties from the percept sequence Information about the way the world evolves Utility information on desirability of world states Action information Goals describe the maximum achievement

Machine Learning methods Symbol-based – symbols represent entities & relationships Connectionist – patterns of activity in networks Genetic – Imitation of genetic & evolutionary process Stochastic – Insight that support Bayes' rule

Exercise

How to locate the license plate area and recognize the license plate info?

Why machine learning? Recent progress in algorithms & theory Growing flood of online data Computational power is available Budding industry

Three niches on machine learning Data mining: using historical data to improve decision Software applications we can't program by hand – Autonomous driving – Speech recognition Self customizing programs - Newsreader that learns user interests

Example of data mining Credit risk analysis Customer purchase behaviour Medical data

Example of too difficult to program by hand Autonomous car – ALVINN drives 70 mph on highways

Example of software that customize to users

Where is this headed? Today: tip of the iceberg First generation algorithms: neural nets, decision trees, regression Applied to well-formated database Budding industry

Where is this headed? (cont.) Opportunity for tomorrow: enormous impact Learn across full mixed-media data Learn across multiple internal databases, plus the web and newsfeeds Learn by active experimentation Learn decisions rather than predictions Cumulative, lifelong learning Programming languages with learning embedded

Relevant disciplines AI Bayesian methods Computational complexity theory Control theory Information theory Philosophy Psychology & neurobiology Statistics....

What is the learning problem? Learning = Improving with experience at some task Improve over task T, With respect to performance measure P, Based on experience E, Eg. Learn to play checkers – T: play checkers – P: % of games won in world tournament – E: Opportunity to play against self

Learning to play checkers T: play checkers P: percent of games won in world tournament What experience? What exactly should be learn? How shall it be represented? What specific algorithm to learn it?

Type of training experience Direct or indirect? Teacher or not? A problem: is training experience representative of performance goal?

Choose the target function ChooseMove:Board-> Move?? V:Board-> R??...

Possible definition for target function V - If b is a final board state that is won, then v(b) = 100 -If b is a final board state that is lost, then v(b) = If b is a final board state that is drawn, then v(b) = 0 -If b is not a final state in the game, then v(b) = v(b'), where b' is the best final board state that can be archived starting from b and playing optimally until the end of game. This gives correct value but is not operational.

Environmental Friendly Choose representation for target function Collection of rules? Neural networks? Polynomial functions or board features?....

Some issues in machine learning What algorithms can approximate functions well (and when)? How does number of training examples influence accuracy? How does complexity of hypothesis representation impact it? How does noisy data influence accuracy? What are the theoretical limits of learnability?

Some issues in machine learning (cont.) How can prior knowledge of learner help? What clues can we get from biological learning systems? How can systems alter their own representations?

Paradigms in machine learning The simplest form of learning is just assigning values to specified parameters. This is a situation when the system contains all the knowledge required for a particular type of Tasks. Another rudimentary type of learning is storing data as it is. This is called rote learning. An example of this type of learning is filling a database.

Paradigms in machine learning The process of knowledge acquisition in an expert system is a kind of learning task where some pre-defined structures (rules, frames etc.) are filled with data specified directly or indirectly by an expert. In this case only the structure of the knowledge is known. The system is given a set of examples (training data) and it is supposed to create a description of this set in terms of a particular language with given priori knowledge and some characteristics of the domain. This is a typical task for Inductive learning and is usually called Concept learning or Learning from examples.

Paradigms in machine learning There are learning systems (e.g.Neural networks) which given no priori knowledge can learn to react properly to the data. Neural networks actually use a kind of a pre-defined structure of the knowledge to be represented (a network of neuron-like elements), which however is very general and thus suitable for various kinds of knowledge. Reference: Lecture Notes in Machine Learning Zdravko Markov, 2003

General steps in ML Problem statement Feature extraction ML implementation Performance evaluation

Example problem: movie critics Example method: K-means clustering algorithm

Group discussion Group 1 – amira, meng kwang, nogol, musobi, hadi Problem: ML for recycling centre – paper, glass, plastic

Group discussion Group 2 – airil, bryan,afsaneh2,farzaneh, azleen Problem: ML for junk filter

Group discussion Group 3 – malina, syaza, laith, afsaneh1, mehrdad Problem: ML for electronics gadget recommendation system ( can be specific – smartphone or tablet)

Group discussion – expected discussion 1. Explanation on problem statement 2. List of appropriate features 3. Chosen features (significant) 4. ML Method 5. Expected results

Class discussion 1. How do algorithms make recommendations from data? 2. Why are features important? 3. Would K-means work the same with more than 2 features? 4. Could we visualize more than 2 features? More than 3? 5. Think of how Euclidean Distance is calculated. Do all the features need to be on the same scale? 6. What are challenges in solving your problem?

Case study: W5-W6 Discuss on what are challenges (at least 3) that AI has to face in your research problem. Then, propose a solution that can overcome the challenges. Submission date: 21 October 2011 (Friday) at 2pm via . Maximum page: 3 (inc. reference/figure/table), font size 12, spacing 1.5.