Artificial Intelligence for Business MIS 272 Lecture Notes PDF

Artificial Intelligence for Business MIS 272 Module 2: Technologies Lesson 7: Algorithms Students Learning Outcomes 6.1 Types of Problems 6.2 Types of Machine Learning Algorithms 6.3 Machine Learning Algorithms Types of Problems Problems categories: Classifi...

Artificial Intelligence for Business MIS 272 Module 2: Technologies Lesson 7: Algorithms Students Learning Outcomes 6.1 Types of Problems 6.2 Types of Machine Learning Algorithms 6.3 Machine Learning Algorithms Types of Problems Problems categories: Classification: Figure out in which group a new input belongs. Subdivided into binary and multiclass classification. Binary classification, the machine has only two classes, (yes and no) Multi-class classification, the machine has more than two classes, such as whether a picture contains a car, building, dog... Regression: Approximation based on the data input. For example, predicts stock prices. Types of Problems Clustering: The machine segregates data with similar traits and moves them into clusters. Data is grouped based on similarity. Supervised learning (classification, Regression) Unsupervised learning (clustering) Types of Problems Types of Problems An Algorithm is “a set of rules for solving a problem in a fixed number of steps” Some common used algorithms: decision trees k-nearest neighbor näive Bayes regression analysis k-mean clustering Types of Problems Binary classification: Have only two possible outcomes. Is the hotel room going to be booked next week or not? Will the stock market go up this afternoon or not? Does this picture contain a hot dog or not? Use decision trees or näive Bayes. Types of Problems Multiclass classification: Have three or more classes an image of a food classified as meal, drink, or dessert. Use various statistical algorithms to assign inputs to classes. Use k-nearest neighbor (kNN), näive Bayes, and decision trees. Types of Problems Regression: a value along a continuous range of values Predict the number of rooms that will be booked on a given day. Have no predefined outcomes or classes. The answer will be a value in a continuous range of possibilities.. Machine Learning Algorithms Decision Trees “a flow chart for choosing a course of action or drawing a conclusion.” Used for supervised learning binary classification problems. Specify a number of predictors that determine a given outcome. Plenty of graphics programs for creating decision trees. Decision Tree Decision Tree Machine Learning Algorithms K-nearest neighbor (kNN) Classifies data based on similarities The k represents the number of nearest neighbors. k = 1 means only 1 nearest neighbor. k = 2 means two nearest neighbors. The higher the k value, the broader the category or class. Called instance-based learning (lazy learning). Machine Learning Algorithms Identifying a dog: 1. Choose two key characteristics that serve as predictors, say weight and hair length. 2. Chart your data on a graph with Weight (x) and a Hair (y) axis. 3. Create a training data set based on 1,000 dogs. 4. Measure a new dog’s hair length and weigh. 5. Plot a point on your chart for the new dog. 6. Let’s say that for this data set we use k = 5. 7. Draw a circle around the points you plotted for your new dog and its five closest neighbors. 8. Count the votes obtained for each class of these nearest neighbors. 9. Choose the class with the largest number of votes kNN K=5 Machine Learning Algorithms K-means clustering Commonly confused with k-nearest neighbor (kNN). An unsupervised ML algorithm. K represents the number of groups you want the machine to create. Assigns new data to the group with nearest centroid. Also called instance-based learning (lazy learning). K-means K=3 Machine Learning Algorithms Regression analysis looks at the relationship between predictors and outcomes to make predictions of future outcomes. Predictors are also referred to as: input variables independent variables regressors Feed training data that contains a small collection of predictors and their associated known outcomes The machine develops a model that describes the relationship between predictors and outcomes. Machine Learning Algorithms Regression Analysis Feed test data into the machine The machine tries repeatedly to predict the outcomes This approach is a type of supervised learning. A good model can tell the likely outcome. Linear regression is one of the most common types of ML regression algorithms  Linear regression for ice cream sales vs temp Machine Learning Algorithms Considerations for Regression Analysis More data increases the accuracy of predictions. There’s some debate about whether it’s actually ML is not actually learning anything new about the data. using the data to create a standard statistical model. The key to success is to find the right predictors and to look for some type of linear connection with the outcome. Artificial Intelligence for Business MIS 272 Module 2: Technologies Lesson 8: Deep Learning Students Learning Outcomes 7.1 What is Deep Learning (DL)? 7.2 Artificial Neural Network (ANN) Algorithm 7.3 Variations of ANNs 7.4 Reasons to use ANN and its variations What is Deep Learning? Deep learning is “a broad family of techniques for machine learning in which hypotheses Deep learning take the form of complex algebraic circuits with tunable connection strengths.” The word “deep” refers to the fact that the circuits are typically organized into many layers The most widely used approach for applications such as visual object recognition, machine translation, speech recognition, speech synthesis, and image synthesis plays a significant role in reinforcement learning applications What is Deep Learning? Deep Learning (DL) Tried to model networks of neurons in the brain (McCulloch and Pitts, 1943) with computational circuits. Often called neural networks Resemblance to real neural cells and structures is superficial. What is Deep Learning? The Anatomy of Artificial Intelligence What is Deep Learning? The true reasons for the success of deep learning have yet to be fully elucidated. Has self-evident advantages over some of the methods for high- dimensional data such as images. Linear Regression Decision List Artificial Neural Network Linear Regression Vs Decision List Vs Deep Learning Artificial Neural Network ◌Artificial ِ Neural Network (ANN) is inspired by the human brain. the same way a paper airplane is inspired by a supersonic jet. The brain is composed of physical entities called neurons An artificial neural network is composed of nodes. Each node accepts and combines weighted input, adds bias (an adjustment), and performs a function on the result to produce an output. That output can then be transferred to other nodes for additional processing. Artificial Neural Network ◌Each ِ node can connect with many other nodes in the network. Nodes are arranged in three layers: input, hidden, and output Signals pass through the network from the input layer, through the hidden layer, to the output layer. The hidden layer may comprise one or numerous layers does the heavy lifting performs the complex calculations. Artificial Neural Network Artificial Neural Network Artificial Neural Network Neurons are processing units (processing elements [PEs]) Perform a set of predefined mathematical operations on the numerical values coming from the input variables from the other neuron outputs create and push out its own outputs. Each input goes into the neuron with an adjustable weight and a bias term. A multiplication weight function applies the weight to the input A net input function adds the bias term to the weighted input. The output of the net input function goes through transfer (aka activation) function for conversion and the production of the actual output. Artificial Neural Network Weight allow nodes to determine which inputs to pay more attention to. Bias (a positive or negative value) can increase or decrease the value of the output. Artificial Neural Network Some of the most common transfer functions: Hard Limit Linear Sigmoid Selection of proper transfer functions: a broad knowledge of neural networks characteristics of the data Purpose of the artificial neural network Variation of Artificial Neural Network Most applications involved network architectures with only a few hidden layers and a limited number of neurons. Suffer from limitation of central processing units (CPU) Graphics processing units (GPUs) technology has enabled us to successfully run neural networks with over a million neurons. Able to go deeper into the data features and extract more sophisticated patterns Variation of Artificial Neural Network Multilayer Perceptron (MLP) Also known as deep feedforward networks The most general type of deep networks large-scale neural networks that can contain many layers of neurons The flow of information that goes through them is always forwarding Recurrent neural networks (RNN). Feedback connections are allowed Variation of Artificial Neural Network Convolutional Neural Network (CNN) Among the most popular types of deep learning methods. A variations of the deep MLP architecture Designed for computer vision applications image processing video processing text recognition applicable to nonimage data sets Why Artificial Neural Network? Why scientists were inspired by biological neurons? The human brain is extremely efficient at classification. Can quickly distinguish different handwriting styles, ethnic foods, and species of animals. An artificial neural network is most effective with large data sets. The more quality data it has, the more accurately it can predict outputs. Why Artificial Neural Network? ANN can allow the machine to learn through any of the following approaches: Supervised learning Unsupervised learning Semi-supervised learning Reinforcement learning