Introducing Machine Learning - MathWorks

IntroducingMachineLearning

What is MachineLearning?Machine learning teaches computers to do what comes naturally tohumans and animals: learn from experience. Machine learning algorithmsuse computational methods to “learn” information directly from datawithout relying on a predetermined equation as a model. The algorithmsadaptively improve their performance as the number of samples availablefor learning increases.

More Data,More Questions,Better AnswersReal-World ApplicationsWith the rise in big data, machine learning hasbecome particularly important for solving problemsin areas like these: scoring and algorithmic trading Machine learning algorithms find natural patterns in datathat generate insight and help you make better decisions andpredictions. They are used every day to make critical decisionsin medical diagnosis, stock trading, energy load forecasting, andmore. Media sites rely on machine learning to sift through millionsof options to give you song or movie recommendations. Retailersuse it to gain insight into their customers’ purchasing behavior.Computational finance, for creditImage processing and computer vision,for face recognition, motion detection, andobject detection Computational biology, for tumordetection, drug discovery, and DNAsequencing Energy production, for price and loadforecasting Automotive, aerospace, andmanufacturing, for predictive maintenance 3Introducing Machine LearningNatural language processing

How MachineLearning WorksMachine learning uses two types of techniques: supervisedlearning, which trains a model on known input and output data sothat it can predict future outputs, and unsupervised learning, whichfinds hidden patterns or intrinsic structures in input data.Machine Learning TechniquesUNSUPERVISEDLEARNINGGroup and interpretdata based onlyon input dataMACHINE LEARNINGSUPERVISEDLEARNINGDevelop predictivemodel based on bothinput and output data4Introducing Machine LearningCLUSTERINGCLASSIFICATIONREGRESSION

SupervisedLearningThe aim of supervised machine learning is to build a modelthat makes predictions based on evidence in the presence ofuncertainty. A supervised learning algorithm takes a known set ofinput data and known responses to the data (output) and trains amodel to generate reasonable predictions for the responseto new data.Supervised learning uses classification and regression techniquesto develop predictive models.5 Classification techniques predict discrete responses—forexample, whether an email is genuine or spam, or whethera tumor is cancerous or benign. Classification modelsclassify input data into categories. Typical applicationsinclude medical imaging, speech recognition, andcredit scoring. Regression techniques predict continuous responses—for example, changes in temperature or fluctuations inpower demand. Typical applications includeelectricity load forecasting and algorithmic trading.Introducing Machine LearningUsing Supervised Learning toPredict Heart AttacksSuppose clinicians want to predict whethersomeone will have a heart attack within a year.They have data on previous patients, including age,weight, height, and blood pressure. They knowwhether the previous patients had heart attackswithin a year. So the problem is combining theexisting data into a model that can predict whethera new person will have a heart attackwithin a year.

UnsupervisedLearningUnsupervised learning finds hidden patterns or intrinsic structuresin data. It is used to draw inferences from datasets consisting ofinput data without labeled responses.Clustering is the most common unsupervised learningtechnique. It is used for exploratory data analysis to find hiddenpatterns or groupings in data.Applications for clustering include gene sequence analysis,market research, and object recognition.6Introducing Machine LearningClusteringPatterns inthe Data

How Do You DecideWhich Algorithmto Use?Selecting an AlgorithmMACHINE LEARNINGChoosing the right algorithm can seem overwhelming—thereare dozens of supervised and unsupervised machine learningalgorithms, and each takes a different approach to learning.There is no best method or one size fits all. Finding the rightalgorithm is partly just trial and error—even highly experienceddata scientists can’t tell whether an algorithm will work withouttrying it out. But algorithm selection also depends on the size andtype of data you’re working with, the insights you want to get fromthe data, and how those insights will be used.7Introducing Machine SIFICATIONREGRESSIONCLUSTERINGSupport VectorMachinesLinear Regression,GLMK-Means, K-MedoidsFuzzy C-MeansDiscriminantAnalysisSVR, GPRHierarchicalNaive BayesEnsemble MethodsGaussian MixtureNearest NeighborDecision TreesNeural NetworksNeural NetworksHidden MarkovModel

When ShouldYou Use MachineLearning?Consider using machine learning when you have a complex task orproblem involving a large amount of data and lots of variables, butno existing formula or equation. For example, machine learning is agood option if you need to handle situations like these:Hand-written rules and equationsare too complex—as in facerecognition and speech recognition.8Introducing Machine LearningThe rules of a task are constantlychanging—as in fraud detectionfrom transaction records.The nature of the data keepschanging, and the program needsto adapt—as in automated trading,energy demand forecasting, andpredicting shopping trends.

Real-World ExamplesCreating Algorithms that Can AnalyzeWorks of ArtResearchers at the Art and Artificial IntelligenceLaboratory at Rutgers University wanted to see whethera computer algorithm could classify paintings by style,genre, and artist as easily as a human. They began byidentifying visual features for classifying a painting’sstyle. The algorithms they developed classified thestyles of paintings in the database with 60% accuracy,outperforming typical non-expert humans.The researchers hypothesized that visual features usefulfor style classification (a supervised learning problem)could also be used to determine artistic influences (anunsupervised problem).They used classification algorithms trained on Googleimages to identify specific objects. They tested thealgorithms on more than 1,700 paintings from 66different artists working over a span of 550 years. Thealgorithm readily identified connected works, includingthe influence of Diego Velazquez’s “Portrait of PopeInnocent X” on Francis Bacon’s “Study After Velazquez’sPortrait of Pope Innocent X.”9Introducing Machine Learning

Real-World ExamplesOptimizing HVAC Energy Usage inLarge BuildingsThe heating, ventilation, and air-conditioning (HVAC)systems in office buildings, hospitals, and other largescale commercial buildings are often inefficient becausethey do not take into account changing weather patterns,variable energy costs, or the building’s thermal properties.Building IQ’s cloud-based software platform addressesthis problem. The platform uses advanced algorithmsand machine learning methods to continuouslyprocess gigabytes of information from power meters,thermometers, and HVAC pressure sensors, as well asweather and energy cost. In particular, machine learningis used to segment data and determine the relativecontributions of gas, electric, steam, and solar powerto heating and cooling processes. The building IQplatform reduces HVAC energy consumption in largescale commercial buildings by 10% - 25% during normaloperation.10 Introducing Machine Learning

Real-World ExamplesDetecting Low-Speed Car CrashesWith more than 8 million members, the RAC is one of theUK’s largest motoring organizations, providing roadsideassistance, insurance, and other services to private andbusiness motorists.To enable rapid response to roadside incidents,reduce crashes, and mitigate insurance costs, the RACdeveloped an onboard crash sensing system that usesadvanced machine learning algorithms to detect lowspeed collisions and distinguish these events from morecommon driving events, such as driving over speedbumps or potholes. Independent tests showed the RACsystem to be 92% accurate in detecting test crashes.11 Introducing Machine Learning

Learn MoreReady for a deeper dive? Explore these resources to learn more aboutmachine learning methods, examples, and tools.WatchMachine Learning Made Easy 34:34Signal Processing and Machine Learning Techniques for Sensor Data Analytics 42:45ReadMachine Learning Blog Posts: Social Network Analysis, Text Mining, Bayesian Reasoning, and moreThe Netflix Prize and Production Machine Learning Systems: An Insider LookMachine Learning Challenges: Choosing the Best Model and Avoiding OverfittingExploreMATLAB Machine Learning ExamplesMachine Learning SolutionsClassify Data with the Classification Learner App 2016 The MathWorks, Inc. MATLAB and Simulink are registered trademarks of The MathWorks, Inc. See mathworks.com/trademarks for a list of additional trademarks.Other product or brand names may be trademarks or registered trademarks of their respective holders.92991v00

object detection Computational biology, for tumor detection, drug discovery, and DNA . So the problem is combining the existing data into a model that can predict whether a new person will have a heart attack within a year. 6 ntroducing Machine Learning Unsupervised Learning Unsupervised learning finds hidden patterns or intrinsic structures in data. It is used to draw inferences from .