Week 03:
	-
	-

---

Topic: Supervised Learning II

---

Objectives:

• Define learning in the context of artificial intelligence
• Define supervised and unsupervised learning and identify when supervised and unsupervised learning should be used
• Define and give examples of regression and classification in supervised learning and identify when regression or classification should be used
• Describe and use various supervised and unsupervised learning algorithms and specify the model characteristics in those algorithms:
  • Global versus Local
  • Hierarchical versus Non-Hierarchical
  • Parametric versus Non-Parametric

---

1. Study the following parts of the Open Source Computer Vision (OpenCV) Version 2.4.4 Application Programming Interface (API):
   Note: OpenCV version 2.4.4 is only located on linprog4.cs.fsu.edu. In order to compile, an OpenCV 2.4.4 C++11 program in main.cpp using g++ version 4.7.2, use the following command on linprog4.cs.fsu.edu:
   g++47 -o main.exe main.cpp -std=c++11 -O3 -Wall -Wextra -Werror -I. -I/usr/local/include/ -I/usr/local/include/boost_1.53.0/ -L/usr/local/lib64/ -lopencv_calib3d -lopencv_contrib -lopencv_core -lopencv_features2d -lopencv_flann -lopencv_gpu -lopencv_highgui -lopencv_imgproc -lopencv_legacy -lopencv_ml -lopencv_nonfree -lopencv_objdetect -lopencv_photo -lopencv_stitching -lopencv_ts -lopencv_video -lopencv_videostab -Wl,-rpath,/usr/local/lib64/ && ./main.exe
   • Introduction
   • API Concepts
     • cv Namespace
     • Automatic Memory Management
     • Fixed Pixel Types. Limited Use of Templates
     • Error Handling
   • Basic Structures
     Note: A vector is a N x 1 or a 1 x N matrix.
     • Mat_ ... the class that you will usually use. If you use this class cv::Mat_<double>, then you don't have to specify the type using the CV_64F OpenCV macro.
     • Mat ... the class that cv::Mat_<T> derives from.
     • Mat::Mat: Various Mat constructors
     • Mat::zeros: Returns a zero array of the specified size and type.
     • Mat::ones: Returns an array of all 1’s of the specified size and type.
     • Mat::eye: Returns an identity matrix of the specified size and type.
     • Mat::diag: Extracts a diagonal from a matrix, or creates a diagonal matrix.
     • Mat::create: Allocates new array data if needed.
     • Matrix Expressions
     • Mat::dot: Computes a dot-product of two vectors.
     • Mat::t: Transposes a matrix.
     • Mat::inv: Inverses a matrix.
     • Mat::inv( DECOMP_SVD ): If the matrix is singular or even non-square, the pseudo inversion is computed.
   • Operations on Arrays
     • determinant: Returns the determinant of a square floating-point matrix.
     • trace: Returns the trace of a matrix.
     • norm: Calculates an absolute array norm, an absolute difference norm, or a relative difference norm.
     • normalize: Normalizes the norm or value range of an array.
     • mean: Calculates an average (mean) of array elements.
     • calcCovarMatrix: Calculates the covariance matrix of a set of vectors.
     • SVD: Class for computing Singular Value Decomposition of a floating-point matrix. The Singular Value Decomposition is used to solve least-square problems, under-determined linear systems, invert matrices, compute condition numbers, and so on.
     • SVD::compute: Performs SVD of a matrix
     • sqrt: Calculates a square root of array elements.
     • min: Calculates per-element minimum of two arrays or an array and a scalar.
     • max: Calculates per-element maximum of two arrays or an array and a scalar.
     • sum: Calculates the sum of array elements.
     • reduce: Reduces a matrix to a vector.
   • ml. Machine Learning
     • MLData
       • CvMLData: Class for loading the data from a .csv file.
       • CvMLData::read_csv: Reads the data set from a .csv-like filename file and stores all read values in a matrix.
       • CvMLData::set_response_idx: Specifies index of response column in the data matrix.
       • CvMLData::change_var_idx: Enables or disables particular variable in the loaded data.
       • CvMLData::get_class_labels_map: Returns a map that converts strings to labels.
       • CvTrainTestSplit: Structure setting the split of a data set read by CvMLData.
       • CvMLData::set_train_test_split: Divides the read data set into two disjoint training and test subsets.
       • CvMLData::get_train_sample_idx: Returns the matrix of sample indices for a training subset.
       • CvMLData::get_test_sample_idx: Returns the matrix of sample indices for a testing subset.
       • CvMLData::get_var_idx: Returns the indices of the active variables in the data matrix.
       • CvMLData::get_values: Returns a pointer to the matrix of predictors and response values.
     • K-Nearest Neighbors
       • CvKNearest: The class implements K-Nearest Neighbors model as described in the beginning of this section.
       • CvKNearest::CvKNearest: Default and training constructors.
       • CvKNearest::train: Trains the model.
       • CvKNearest::find_nearest: Finds the neighbors and predicts responses for input vectors.
   
   
2. Watch the "IV. Linear Regression with Multiple Variables (Week 2)", "VI. Logistic Regression (Week 3)", and "VII. Regularization (Week 3)" series of videos in Machine Learning with Andrew Ng course from Coursera:
   1. Sign In to Coursera.
   2. Click the Explore Courses link at the top of the page.
   3. Search for Machine Learning using the search text box.
   4. Select the Machine Learning with Andrew Ng.
   5. Press the "Preview" button.
   6. Watch the following videos in the "IV. Linear Regression with Multiple Variables (Week 2)" section (61 min):
      • Multiple Features (8 min)
      • Gradient Descent for Multiple Variables (5 min)
      • Gradient Descent in Practice I - Feature Scaling (9 min)
      • Gradient Descent in Practice II - Learning Rate (9 min)
      • Features and Polynomial Regression (8 min)
      • Normal Equation (16 min)
      • Normal Equation Noninvertibility (Optional) (6 min)
   7. Watch the following videos in the "VI. Logistic Regression (Week 3)" section (71 min):
      • Classification (8 min)
      • Hypothesis Representation (7 min)
      • Decision Boundary (15 min)
      • Cost Function (11 min)
      • Simplified Cost Function and Gradient Descent (10 min)
      • Advanced Optimization (14 min)
      • Multiclass Classification: One-vs-all (6 min)
   8. Watch the following videos in the "VII. Regularization (Week 3)" section (40 min):
      • The Problem of Overfitting (10 min)
      • Cost Function (10 min)
      • Regularized Linear Regression (11 min)
      • Regularized Logistic Regression (9 min)
   Reference: Coursera's Machine Learning course by Andrew Ng.
   
   
3. Watch the "Lecture1", "Lecture2", and "Lecture3" series of videos in Neural Networks for Machine Learning with Geoffrey Hinton course from Coursera:
   1. Sign In to Coursera.
   2. Click the Explore Courses link at the top of the page.
   3. Search for Neural Networks for Machine Learning using the search text box.
   4. Select the Neural Networks for Machine Learning with Geoffrey Hinton.
   5. Press the "Sign Up" button and then the "Go to class" button ... or click the "View class archive" link (whichever you are presented with).
   6. Select the Video Lectures menu option on the left hand side on the page.
   7. Watch the following videos in the "Lecture1" section (43 min):
      • Why do we need machine learning? [13 min]
      • What are neural networks? [8 min]
      • Some simple models of neurons [8 min]
      • A simple example of learning [6 min]
      • Three types of learning [8 min]
   8. Watch the following videos in the "Lecture2" section (41 min):
      • Types of neural network architectures [7 min]
      • Perceptrons: The first generation of neural networks [8 min]
      • A geometrical view of perceptrons [6 min]
      • Why the learning works [5 min]
      • What perceptrons can't do [15 min]
   9. Watch the following videos in the "Lecture3" section (43 min):
      • Learning the weights of a linear neuron [12 min]
      • The error surface for a linear neuron [5 min]
      • Learning the weights of a logistic output neuron [4 min]
      • The backpropagation algorithm [12 min]
      • Using the derivatives computed by backpropagation [10 min]
   Reference: Coursera's Neural Networks for Machine Learning course by Geoffrey Hinton.
   
   
4. Read the following pages from Artificial Intelligence: A Modern Approach, Third Edition (cited as AIMA):
   • Required: 50 Pages

     Order	Topic	Book	Section	Page Numbers	Pages	Presentations
     1.	Multivariate Linear Regression	AIMA	Section 18.6.2:	Pages 720–723	(4 Pages)	Learning from Observations and Machine Learning
     2.	Linear Classification with Logistic Regression	AIMA	Section 18.6.4:	Pages 725–727	(3 Pages)	Learning from Observations and Machine Learning
     3.	Artificial Neural Networks	AIMA	Section 18.7:	Pages 727–737	(11 Pages)	Neural Networks and Machine Learning
     4.	Support Vector Machines	AIMA	Section 18.9:	Pages 744–748	(5 Pages)	Learning from Observations and Machine Learning
     5.	Intro to Learning Probabilistic Models	AIMA	Section 20 Intro:	Page 802	(1 Page)	Statistical Learning and Machine Learning
     6.	Statistical Learning	AIMA	Section 20.1:	Pages 802–805	(4 Pages)	Statistical Learning and Machine Learning
     7.	Intro to Learning with Complete Data	AIMA	Section 20.2 Intro:	Page 806	(1 Page)	Statistical Learning and Machine Learning
     8.	Maximum-Likelihood Parameter Learning: Discrete Models	AIMA	Section 20.2.1:	Pages 806–808	(3 Pages)	Statistical Learning and Machine Learning
     9.	Naive Bayes Models	AIMA	Section 20.2.2:	Pages 808–809	(2 Pages)	Statistical Learning and Machine Learning
     10.	Maximum-Likelihood Parameter Learning: Continuous Models	AIMA	Section 20.2.3:	Pages 809–810	(2 Pages)	Statistical Learning and Machine Learning
     11.	Bayesian Parameter Learning	AIMA	Section 20.2.4:	Pages 810–813	(4 Pages)	Statistical Learning and Machine Learning
     12.	Learning Bayes Net Structures	AIMA	Section 20.2.5:	Pages 813–814	(2 Pages)	Statistical Learning and Machine Learning
     13.	Intro to Nonparametric Models	AIMA	Section 18.8 Intro:	Page 737	(1 Page)	Learning from Observations and Machine Learning
     14.	Nearest Neighbor Models	AIMA	Section 18.8.1:	Pages 738–739	(2 Pages)	Learning from Observations and Machine Learning
     15.	Finding Nearest Neighbors with k-d Trees	AIMA	Section 18.8.2:	Pages 739–740	(2 Pages)	Learning from Observations and Machine Learning
     16.	Density Estimation with Nonparametric Models	AIMA	Section 20.2.6:	Pages 814–816	(3 Pages)	Statistical Learning and Machine Learning

     
     
   Reference: University of California, Berkeley's CS 188: Introduction to Artificial Intelligence (Fall 2005) by Stuart Russell
   Reference: Stanford University's CS221: Artificial Intelligence: Principles and Techniques (Autumn 2012-2013) by Percy Liang
   
   
5. Do Assignment 02.