Amit Sethi, EEE, IIT Cepstrum, Oct 16, 2011 1.

Slides:



Advertisements
Similar presentations
Pat Langley Computational Learning Laboratory Center for the Study of Language and Information Stanford University, Stanford, California
Advertisements

Introduction to Machine Learning BITS C464/BITS F464
Godfather to the Singularity
Data Mining Classification: Alternative Techniques
Machine learning continued Image source:
Computational Methods for Data Analysis
An Overview of Machine Learning
Introduction to Machine Learning Algorithms. 2 What is Artificial Intelligence (AI)? Design and study of computer programs that behave intelligently.
ETHEM ALPAYDIN © The MIT Press, Lecture Slides for.
1 MACHINE LEARNING TECHNIQUES IN IMAGE PROCESSING By Kaan Tariman M.S. in Computer Science CSCI 8810 Course Project.
Learning Programs Danielle and Joseph Bennett (and Lorelei) 4 December 2007.
Statistical Learning: Pattern Classification, Prediction, and Control Peter Bartlett August 2002, UC Berkeley CIS.
Part I: Classification and Bayesian Learning
Introduction to machine learning
Machine Learning Usman Roshan Dept. of Computer Science NJIT.
CS 391L: Machine Learning Introduction
CS Machine Learning. What is Machine Learning? Adapt to / learn from data  To optimize a performance function Can be used to:  Extract knowledge.
CSC 4510 – Machine Learning Dr. Mary-Angela Papalaskari Department of Computing Sciences Villanova University Course website:
Artificial Intelligence (AI) Addition to the lecture 11.
Lecture 2: Introduction to Machine Learning
Data Mining Joyeeta Dutta-Moscato July 10, Wherever we have large amounts of data, we have the need for building systems capable of learning information.
MACHINE LEARNING 張銘軒 譚恆力 1. OUTLINE OVERVIEW HOW DOSE THE MACHINE “ LEARN ” ? ADVANTAGE OF MACHINE LEARNING ALGORITHM TYPES  SUPERVISED.
Mehdi Ghayoumi Kent State University Computer Science Department Summer 2015 Exposition on Cyber Infrastructure and Big Data.
Machine Learning1 Machine Learning: Summary Greg Grudic CSCI-4830.
1 CS 512 Machine Learning Berrin Yanikoglu Slides are expanded from the Machine Learning-Mitchell book slides Some of the extra slides thanks to T. Jaakkola,
IE 585 Introduction to Neural Networks. 2 Modeling Continuum Unarticulated Wisdom Articulated Qualitative Models Theoretic (First Principles) Models Empirical.
Machine Learning An Introduction. What is Learning?  Herbert Simon: “Learning is any process by which a system improves performance from experience.”
Bayesian networks Classification, segmentation, time series prediction and more. Website: Twitter:
Introduction to machine learning and data mining 1 iCSC2014, Juan López González, University of Oviedo Introduction to machine learning Juan López González.
1 SUPPORT VECTOR MACHINES İsmail GÜNEŞ. 2 What is SVM? A new generation learning system. A new generation learning system. Based on recent advances in.
1 Mining in geographic data Original slides:Raymond J. Mooney University of Texas at Austin.
Lecture 10: 8/6/1435 Machine Learning Lecturer/ Kawther Abas 363CS – Artificial Intelligence.
Machine Learning Using Support Vector Machines (Paper Review) Presented to: Prof. Dr. Mohamed Batouche Prepared By: Asma B. Al-Saleh Amani A. Al-Ajlan.
Machine Learning.
Machine Learning Tutorial Amit Gruber The Hebrew University of Jerusalem.
1 Machine Learning 1.Where does machine learning fit in computer science? 2.What is machine learning? 3.Where can machine learning be applied? 4.Should.
Learning from observations
I Robot.
Machine Learning Extract from various presentations: University of Nebraska, Scott, Freund, Domingo, Hong,
Machine Learning Introduction. Class Info Office Hours –Monday:11:30 – 1:00 –Wednesday:10:00 – 1:00 –Thursday:11:30 – 1:00 Course Text –Tom Mitchell:
يادگيري ماشين Machine Learning Lecturer: A. Rabiee
Data Mining and Decision Support
1 Introduction to Machine Learning Chapter 1. cont.
Supervised Machine Learning: Classification Techniques Chaleece Sandberg Chris Bradley Kyle Walsh.
Introduction Machine Learning: Chapter 1. Contents Types of learning Applications of machine learning Disciplines related with machine learning Well-posed.
WHAT IS DATA MINING?  The process of automatically extracting useful information from large amounts of data.  Uses traditional data analysis techniques.
WHAT IS DATA MINING?  The process of automatically extracting useful information from large amounts of data.  Uses traditional data analysis techniques.
SUPERVISED AND UNSUPERVISED LEARNING Presentation by Ege Saygıner CENG 784.
Network Management Lecture 13. MACHINE LEARNING TECHNIQUES 2 Dr. Atiq Ahmed Université de Balouchistan.
Machine Learning Usman Roshan Dept. of Computer Science NJIT.
DATA MINING and VISUALIZATION Instructor: Dr. Matthew Iklé, Adams State University Remote Instructor: Dr. Hong Liu, Embry-Riddle Aeronautical University.
Brief Intro to Machine Learning CS539
Combining Models Foundations of Algorithms and Machine Learning (CS60020), IIT KGP, 2017: Indrajit Bhattacharya.
Data Mining, Machine Learning, Data Analysis, etc. scikit-learn
Machine Learning overview Chapter 18, 21
Machine Learning overview Chapter 18, 21
Eick: Introduction Machine Learning
Intro to Machine Learning
School of Computer Science & Engineering
Introductory Seminar on Research: Fall 2017
CH. 1: Introduction 1.1 What is Machine Learning Example:
Machine Learning Ali Ghodsi Department of Statistics
What is Pattern Recognition?
Basic Intro Tutorial on Machine Learning and Data Mining
Data Warehousing and Data Mining
Overview of Machine Learning
MACHINE LEARNING TECHNIQUES IN IMAGE PROCESSING
MACHINE LEARNING TECHNIQUES IN IMAGE PROCESSING
Data Mining, Machine Learning, Data Analysis, etc. scikit-learn
Data Mining, Machine Learning, Data Analysis, etc. scikit-learn
Presentation transcript:

Amit Sethi, EEE, IIT Cepstrum, Oct 16,

Objectives:  Understand what is machine learning  Motivate why it has become so important  Identify Types of learning and salient frameworks, algorithms and their utility  Take a sneak peak at the next set of problems 2

 What is learning?  Why learn?  Types of learning and salient frameworks  Frontiers 3

 Example: Learning to ride a bicycle  T: Task of learning to ride a bicycle  P: Performance of balancing while moving  E: Experience of riding in many situations  Is it wise to memorize all situations and appropriate responses by observing an expert? 4

Improve on task, T, with respect to performance metric, P, based on experience, E. T: Playing checkers P: Percentage of games won against an arbitrary opponent E: Playing practice games against itself T: Recognizing hand-written words P: Percentage of words correctly classified E: Database of human-labeled images of handwritten words T: Driving on four-lane highways using vision sensors P: Average distance traveled before a human-judged error E: A sequence of images and steering commands recorded while observing a human driver. T: Categorize messages as spam or legitimate. P: Percentage of messages correctly classified. E: Database of s, some with human-given labels Source: Introduction to Machine Learning by Raymond J. Mooney 5

 Determine f such that y n =f(x n ) and g(y, x) is minimized for unseen values of y and x pairs.  Form of f is fixed, but some parameters can be tuned:  So, y=f θ (x ), where, x is observed, and y needs to be inferred  e.g. y=1, if mx > c, 0 otherwise, so θ = (m,c)  Machine Learning is concerned with designing algorithms that learn “better” values of θ given “more” x (and y) for a given problem 6

 What is the scope of the task?  How will performance be measured?  How should learning be approached?  Scalability:  How can we learn fast?  How much resources are needed to learn?  Generalization:  How will it perform in unseen situations?  Online learning:  Can it learn and improve while performing the task? 7

 Artificial Intelligence  Data Mining  Probability and Statistics  Information theory  Numerical optimization  Adaptive Control Theory  Neurobiology  Psychology (cognitive, perceptual, dev.)  Linguistics 8

 What is learning?  Why learn?  Types of learning and salient frameworks  Frontiers 9

 Develop systems that are too difficult/expensive to construct manually because they require specific detailed skills or knowledge tuned to a specific task (knowledge engineering bottleneck).  Develop systems that can automatically adapt and customize themselves to individual users.  Personalized news or mail filter  Personalized tutoring  Discover new knowledge from large databases (data mining).  Market basket analysis (e.g. diapers and beer)  Medical text mining (e.g. migraines to calcium channel blockers to magnesium) Source: Introduction to Machine Learning by Raymond J. Mooney 10

 Computational studies of learning may help us understand learning in humans and other biological organisms.  Hebbian neural learning ▪ “Neurons that fire together, wire together.”  Power law of practice log(# training trials) log(perf. time) Source: Introduction to Machine Learning by Raymond J. Mooney 11

 Many basic effective and efficient algorithms available  Large amounts of data available  Large amounts of computational resources available Source: Introduction to Machine Learning by Raymond J. Mooney 12

Automatic vehicle navigation Road recognition Automatic navigation Speech recognition Speech to text Automated services over the phone Face detection Facebook face tagging suggestions Camera autofocus for portraits 13

 What is learning?  Why learn?  Types of learning and salient frameworks  Frontiers 14

 Remember, y=f θ (x)?  y can be continuous or categorical  y may be known for some x or none at all  f can be simple (e.g. linear) or complex  f can incorporate some knowledge of how x was generated or be blind to the generation  etc… 15

 Supervised learning:  For, y=f θ (x), a set of x i, y i (usually classes) are known  Now predict y j for new x j  Examples:  Two classes of protein with given amino acid sequences  Labeled male and female face images 16

 In a nutshell:  Input is non-linearly transformed by hidden layers usually a “fuzzy” linearly classified combination  Output is a linear combination of the hidden layer  Use when:  Want to model a non-linear function  Labeled data is available  Don’t want to write new s/w  Variations:  Competitive learning for classification  Many more… 17

 In a nutshell:  Learns optimal boundary between two classes (red line)  Use when:  Labeled class data is available  Want to minimize chance of error in the test case  Variations:  Non-linear mapping of the input vectors using “Kernels” 18

 Unsupervised learning:  For, y=f θ (x), only a set of x i are known  Predict y, such that y is simpler than x but retains its essence  Examples:  Clustering (when y is a class label)  Dimensionality reduction (when y is continuous) 19

 In a nutshell:  Grouping a similar objects based on a definition of similarity  That is, intra vs. inter cluster similarity, e.g. distance from center of the cluster  Use when:  Class labels are not available, but you have a desired number of clusters in mind  Variations:  Different similarity measures  Automatic detection of number of clusters  Online clustering 20

 In a nutshell:  High dimensional data, where not all dimensions are independent, e.g. (x 1, x 2, x 3 ), where x 3 =ax 1 +bx 2 +c  Use when:  You want to perform linear dimensionality reduction  Variations:  ICA  Online PCA 21

 In a nutshell:  Learning a lower-dimensional manifold (e.g. surface) close to which the data lies  Use when:  You want to perform non- linear dimensionality reduction  Variations:  SOM 22

 Generative models:  For, y=f θ (x), we have some idea of how x was generated given x and θ  Examples:  HMMs: Given phonemes and {age, gender}, we know how the speech can be generated  Bayesian Networks: Given {gender, age, race} we have some idea of what a face will look like for different emotions 23

 Discriminative Models:  Do not care about how the data was generated  Finding the right features is of prime importance  Followed by finding the right classifier  Examples:  SVM  MLP Source: “Automatic Recognition of Facial Actions in Spontaneous Expressions” by Bartlett et al in Journal of Multimedia, Sep

 What is learning?  Why learn?  Types of learning and salient frameworks  Frontiers 25

 1980s:  Advanced decision tree and rule learning  Explanation-based Learning (EBL)  Learning and planning and problem solving  Utility problem  Analogy  Cognitive architectures  Resurgence of neural networks (connectionism, backpropagation)  Valiant’s PAC Learning Theory  Focus on experimental methodology  1990s  Data mining  Adaptive software agents and web applications  Text learning  Reinforcement learning (RL)  Inductive Logic Programming (ILP)  Ensembles: Bagging, Boosting, and Stacking  Bayes Net learning Source: Introduction to Machine Learning by Raymond J. Mooney 26

 2000s  Support vector machines  Kernel methods  Graphical models  Statistical relational learning  Transfer learning  Sequence labeling  Collective classification and structured outputs  Computer Systems Applications ▪ Compilers ▪ Debugging ▪ Graphics ▪ Security (intrusion, virus, and worm detection)  management  Personalized assistants that learn  Learning in robotics and vision Source: Introduction to Machine Learning by Raymond J. Mooney 27

Bioinformatics Gene expression prediction (just scratched the surface) Automated drug discovery Speech recognition Context recog., e.g. for digital personal assistants (SiRi?) Better than Google translate; imagine visiting Brazil Image and video processing Automatic event detection in video “Seeing” software for the blind 28

Robotics Where is my iRobot? Would you raise a “robot” child and make it learn? Advanced scientific calculations Weather modeling through prediction Vector field or FEM calculation through prediction Who knows… Always in search of new problems 29

 Learning the structure of classifiers  Automatic feature discovery and active learning  Discovering the limits of learning  Information theoretic bounds?  Learning that never ends  Explaining human learning  Computer languages with ML primitives 30 Adapted from: “The Discipline of Machine Learning” by Tom Mitchell, 2006

Thank you! 31

 Inference: Using a system to get the output variable for a given input variable  Learning: Changing parameters according to an algorithm to improve performance  Training: Using machine learning algorithm to learn function parameters based on input and (optionally) output dataset known as “training set”  Validation and Testing: Using inference (without training) to test the performance of the learned system on data  Offline learning: When all training happens prior to testing, and no learning takes place during testing  Online learning: When learning and testing happen for the same data 32