Kansas State University Department of Computing and Information Sciences CIS 890: Special Topics in Intelligent Systems Wednesday, November 15, 2000 Cecil.

Slides:



Advertisements
Similar presentations
1 Classification using instance-based learning. 3 March, 2000Advanced Knowledge Management2 Introduction (lazy vs. eager learning) Notion of similarity.
Advertisements

Machine Learning Instance Based Learning & Case Based Reasoning Exercise Solutions.
Data Mining Classification: Alternative Techniques
K-means method for Signal Compression: Vector Quantization
1 CS 391L: Machine Learning: Instance Based Learning Raymond J. Mooney University of Texas at Austin.
Instance Based Learning
1 Machine Learning: Lecture 7 Instance-Based Learning (IBL) (Based on Chapter 8 of Mitchell T.., Machine Learning, 1997)
Lazy vs. Eager Learning Lazy vs. eager learning
Kansas State University Department of Computing and Information Sciences Laboratory for Knowledge Discovery in Databases (KDD) KDD Group Research Seminar.
1er. Escuela Red ProTIC - Tandil, de Abril, Instance-Based Learning 4.1 Introduction Instance-Based Learning: Local approximation to the.
. Markov Chains as a Learning Tool. 2 Weather: raining today40% rain tomorrow 60% no rain tomorrow not raining today20% rain tomorrow 80% no rain tomorrow.
Classification and Decision Boundaries
Navneet Goyal. Instance Based Learning  Rote Classifier  K- nearest neighbors (K-NN)  Case Based Resoning (CBR)
Instance Based Learning
Università di Milano-Bicocca Laurea Magistrale in Informatica Corso di APPRENDIMENTO E APPROSSIMAZIONE Lezione 8 - Instance based learning Prof. Giancarlo.
K nearest neighbor and Rocchio algorithm
MACHINE LEARNING 9. Nonparametric Methods. Introduction Lecture Notes for E Alpaydın 2004 Introduction to Machine Learning © The MIT Press (V1.1) 2 
RBF Neural Networks x x1 Examples inside circles 1 and 2 are of class +, examples outside both circles are of class – What NN does.
Instance based learning K-Nearest Neighbor Locally weighted regression Radial basis functions.
Chapter 2: Pattern Recognition
Instance Based Learning
Instance-Based Learning
Lazy Learning k-Nearest Neighbour Motivation: availability of large amounts of processing power improves our ability to tune k-NN classifiers.
These slides are based on Tom Mitchell’s book “Machine Learning” Lazy learning vs. eager learning Processing is delayed until a new instance must be classified.
1 Nearest Neighbor Learning Greg Grudic (Notes borrowed from Thomas G. Dietterich and Tom Mitchell) Intro AI.
Aprendizagem baseada em instâncias (K vizinhos mais próximos)
KNN, LVQ, SOM. Instance Based Learning K-Nearest Neighbor Algorithm (LVQ) Learning Vector Quantization (SOM) Self Organizing Maps.
Instance Based Learning Bob Durrant School of Computer Science University of Birmingham (Slides: Dr Ata Kabán) 1.
INSTANCE-BASE LEARNING
CS Instance Based Learning1 Instance Based Learning.
Radial-Basis Function Networks
Module 04: Algorithms Topic 07: Instance-Based Learning
Methods in Medical Image Analysis Statistics of Pattern Recognition: Classification and Clustering Some content provided by Milos Hauskrecht, University.
K Nearest Neighborhood (KNNs)
Basic Data Mining Technique
K Nearest Neighbors Saed Sayad 1www.ismartsoft.com.
Chapter 8 The k-Means Algorithm and Genetic Algorithm.
11/12/2012ISC471 / HCI571 Isabelle Bichindaritz 1 Prediction.
 2003, G.Tecuci, Learning Agents Laboratory 1 Learning Agents Laboratory Computer Science Department George Mason University Prof. Gheorghe Tecuci 9 Instance-Based.
© Tan,Steinbach, Kumar Introduction to Data Mining 4/18/ Statistical Inference (By Michael Jordon) l Bayesian perspective –conditional perspective—inferences.
1 Instance Based Learning Ata Kaban The University of Birmingham.
1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 24 Nov 2, 2005 Nanjing University of Science & Technology.
CpSc 881: Machine Learning Instance Based Learning.
CpSc 810: Machine Learning Instance Based Learning.
Fast Query-Optimized Kernel Machine Classification Via Incremental Approximate Nearest Support Vectors by Dennis DeCoste and Dominic Mazzoni International.
Outline K-Nearest Neighbor algorithm Fuzzy Set theory Classifier Accuracy Measures.
Lazy Learners K-Nearest Neighbor algorithm Fuzzy Set theory Classifier Accuracy Measures.
Kansas State University Department of Computing and Information Sciences CIS 798: Intelligent Systems and Machine Learning Tuesday, November 23, 1999.
KNN Classifier.  Handed an instance you wish to classify  Look around the nearby region to see what other classes are around  Whichever is most common—make.
Kansas State University Department of Computing and Information Sciences CIS 732: Machine Learning and Pattern Recognition Thursday, 05 April 2007 William.
K nearest neighbors algorithm Parallelization on Cuda PROF. VELJKO MILUTINOVIĆ MAŠA KNEŽEVIĆ 3037/2015.
Meta-learning for Algorithm Recommendation Meta-learning for Algorithm Recommendation Background on Local Learning Background on Algorithm Assessment Algorithm.
CS Machine Learning Instance Based Learning (Adapted from various sources)
K-Nearest Neighbor Learning.
Eick: kNN kNN: A Non-parametric Classification and Prediction Technique Goals of this set of transparencies: 1.Introduce kNN---a popular non-parameric.
Debrup Chakraborty Non Parametric Methods Pattern Recognition and Machine Learning.
Instance-Based Learning Evgueni Smirnov. Overview Instance-Based Learning Comparison of Eager and Instance-Based Learning Instance Distances for Instance-Based.
CS 8751 ML & KDDInstance Based Learning1 k-Nearest Neighbor Locally weighted regression Radial basis functions Case-based reasoning Lazy and eager learning.
1 Instance Based Learning Soongsil University Intelligent Systems Lab.
Instance Based Learning
Data Mining: Concepts and Techniques (3rd ed
Instance Based Learning (Adapted from various sources)
K Nearest Neighbor Classification
یادگیری بر پایه نمونه Instance Based Learning Instructor : Saeed Shiry
Instance Based Learning
Chap 8. Instance Based Learning
Advanced Mathematics Hossein Malekinezhad.
Machine Learning: UNIT-4 CHAPTER-1
Nearest Neighbors CSC 576: Data Mining.
Nearest Neighbor Classifiers
Presentation transcript:

Kansas State University Department of Computing and Information Sciences CIS 890: Special Topics in Intelligent Systems Wednesday, November 15, 2000 Cecil P. Schmidt Department of Computer Information Sciences, KSU Instance-Based Learning KDD Group Presentation

Kansas State University Department of Computing and Information Sciences CIS 890: Special Topics in Intelligent Systems Presentation Outline What is Instance Based Learning? k-Nearest Neighbor Learning Distance-Weighted Nearest Neighbor Algorithm Other IBL Methods –Locally Weighted Regression –Radial Basis Functions –Case-Based Reasoning Lazy Versus Eager Learning Research Opportunities in IBL Summary Bibliography

Kansas State University Department of Computing and Information Sciences CIS 890: Special Topics in Intelligent Systems What is Instance Based Learning? Description –Instance Base Learning (IBL) methods initially store the presented training data. –Upon encountering a new query, the set of similar related instances are retrieved and used for classification. Differences between IBL methods and others –local versus global approximation of query –unique approximation to target function per distinct query instance Advantages –complex global approximation becomes simpler local approximation –instance points can be complex, symbolic representations Disadvantages –Nearly all computation takes place at classification time –Classification typically looks at all attributes of the query instance whereas they might not all be important causing misclassification.

Kansas State University Department of Computing and Information Sciences CIS 890: Special Topics in Intelligent Systems k-Nearest Neighbor Learning Most basic of IBL Methods –Assumes all instances correspond to points in the n-dimensional space  n –Nearest neighbors of a query instance are defined in terms of standard Euclidean distance. Distance d definition –let an arbitrary instance x be described by feature vector  a 1 (x), a 2 (x),…a n (x)  where a r (x) denotes the value of the rth attribute of instance x. The distance between x i and x j is d(x i, x j ), where d(x i, x j )  sqrt(  r = 1..n (a r (x i ) - a r (x j )) 2 Distance Calculation Example –find distance between vector  2,2,4  and vector  2,2,2  –sqrt((2-2) 2 + (2-2) 2 + (2-4) 2 ) = 2

Kansas State University Department of Computing and Information Sciences CIS 890: Special Topics in Intelligent Systems k-Nearest Neighbor Learning Algorithm: Discrete-Valued Target Function k-Nearest Neighbor Learning Algorithm: Discrete-Valued Target Function Algorithm Consists of two parts which include a training part and a classification part. Training Algorithm –For each training_example  x, f(x) , add the example to the list train_examples Classification Algorithm –Given a query instance x q to be classified Let x 1 … x k denote the k instances from training_examples that are nearest to x q Return f-hat(x q )  argmax v  V  i=1..k  (v,f(x i )) where  (a,b) = 1 if a = b and  (a,b) = 0 otherwise

Kansas State University Department of Computing and Information Sciences CIS 890: Special Topics in Intelligent Systems k-Nearest Neighbor Learning Algorithm: Discrete- Valued Target Function Classification Example k-Nearest Neighbor Learning Algorithm: Discrete- Valued Target Function Classification Example Classification Example –Assume following training_examples –x 1 =  1,2,7,8,+ , x 2 =  1,3,5,6,-  –Classify x 3 =  1,2,7,6  (Step 1) Compute distance from each point in training examples d(x 1, x 3 ) = 2 d(x 2, x 3 ) = 1 (Step 2) Classify x 3 based on the nearest k points where k = 1 we classify x 3 as - since it is nearest x 2

Kansas State University Department of Computing and Information Sciences CIS 890: Special Topics in Intelligent Systems K-Nearest Neighor: Real-Valued Target Function Used to Approximate continuous-valued target function Calculates mean value of k-nearest training examples rather than their most common value. Replaces final line of discrete-valued target function with the following: –f-hat(x q )  (  i=1..k f(x i ))/k Example –Given training_examples (x 1,1),(x 2,1),(x 3,0) –We wish to classify x using 2-nearest neighbors –d(x 1,x 4 ) = 5, d(x 2,x 4 ) = 2, d(x 3,x 4 ) = 4 –therefore x 4 is nearest x 2 and x 3 –taking the mean we get (1+0)/2 = 0.5

Kansas State University Department of Computing and Information Sciences CIS 890: Special Topics in Intelligent Systems Distance-weighted Nearest Neighbor Algorithm Weights contribution of each k neighbors according to their distance to the query point x q, giving greater weight to closer neighbors Example –weight each neighbor according to inverse square of its distance from x q f-hat(x q )  argmax v  V  i=1..k w i  (v,f(x i )) where w i  1/d(x q,x i ) 2 if x q exactly matches x i then w i = 1 Modification of real-valued target function normalizes contributions of various weights f-hat(x q )   i=1..k w i f(x i )/  i=1..k w i Adding distance weighting, all training examples can be used to influence classification of x q Classifier will run more slowly Referred to as a global method when all training points are used Referred to as a local method when k-nearest training examples are used

Kansas State University Department of Computing and Information Sciences CIS 890: Special Topics in Intelligent Systems Other IBL Methods Locally Weighted Regression –Constructs explicit approximation of f over a local region surrounding x q. –Uses nearby or distance-weighted training examples to form local approximation –local refers to the fact that only data near the query point is used –weighted refers to fact that contribution of each training example is weighted by distance from query point –regression term is used because it is used widely in statistical learning community for approximating real-valued function. Radial Basis Functions –Closely related to distance-weighted regression and ANNs –Provide global approximation to the target function, represented by a linear combination of many local kernel functions (kernel function K is function of distance used to determine the weight of each training example) Case-Based Reasoning –Uses rich symbolic descriptions to represent instances versus real-valued points in an n-dimensional Euclidean space. –Requires more elaborate retrieval of similar instances (nearest neighbors) –Applied to conceptual design problems - similar designs

Kansas State University Department of Computing and Information Sciences CIS 890: Special Topics in Intelligent Systems Lazy Versus Eager Learing Lazy Learing –Defers decision to generalize beyond training data until each new query instance is encountered –examples include k-nearest neighbor, locally-weighted regression, and case-based reasoning Eager Learning –Generalizes beyond the training data before observing the new query –examples include Decision Tree Learning algorithms such as ID3 and ANNs Differences –Computation Time –Lazy methods generally require less computation during training –Lazy methods require more computation time during classification –Classification Differences –Lazy methods may consider the query instance xq when deciding how to generalize beyond the training data D –Eager methods have already chosen their (global) approximation to the target function. –Lazy learner has option of representing target function by a combination of many local approximation, whereas eager methods must commit at training time.

Kansas State University Department of Computing and Information Sciences CIS 890: Special Topics in Intelligent Systems Research Opportunities In IBL Improved methods for indexing instances which may be rich relational descriptions (CBR) Development of eager methods which employ multiple local approximations to achieve similar effects as lazy learning methods but reduce computation at classification time. (RBF learning attempts this) Applications (Can you think of some others?) –Finding similar software patterns in currently implemented software or in analysis or design models. –Matching database query to most likely instances

Kansas State University Department of Computing and Information Sciences CIS 890: Special Topics in Intelligent Systems Summary Nearest neighbor algorithms are examples of IBL methods which delay much of the classification computation until classification time. These algorithms use local approximation versus global approximation which can have the effect of having unique target functions for each query instance. Ability to model complex target functions by a collection of less complex local approximations Information present in training examples is never lost Distance metric can become misleading as all attributes are considered Employ lazy learning versus eager learning techniques

Kansas State University Department of Computing and Information Sciences CIS 890: Special Topics in Intelligent Systems Bibliography Primary material for this presentation came from: –Mitchell, T. (1997) Machine Learning, MIT Press and The McGraw-Hill Companies, Inc.Boston, Mass.