Fundamentals of Machine Learning Bootcamp - 24 Nov London
0 likes370 views
This document provides information about a machine learning bootcamp offered by Persontyle Ltd. The 5-day, hands-on bootcamp will cover fundamental machine learning concepts, techniques, models and methods. Attendees will learn how to apply over 20 machine learning techniques using the R programming language. The schedule outlines 19 modules that will cover topics like regression, classification, clustering, neural networks, and model selection. Successful completion of the bootcamp will provide attendees with practical skills in applying machine learning to solve real-world problems.
Fundamentals of Machine Learning Bootcamp - 24 Nov London
- 1. © 2014 Persontyle Ltd. All rights reserved. FUNDAMENTALS OF MACHINE LEARNING BOOTCAMPHANDS-ON INTRODUCTION TO MACHINE LEARNING MODELS, METHODS AND ALGORITHMS
- 2. www.persontyle.com © 2014 Persontyle Ltd. All rights reserved. “THE FIELD OF MACHINE LEARNING IS CONCERNED WITH THE QUESTION OF HOW TO CONSTRUCT COMPUTER PROGRAMS THAT AUTOMATICALLY IMPROVE WITH EXPERIENCE.” -TOM MITCHELL MACHINELEARNING
- 3. www.persontyle.com © 2014 Persontyle Ltd. All rights reserved. Datageneratedthroughouractivitiescapturesplethoraofinformationaboutouridentity,likesanddislikesetc.Thisinformationhastremendousvalueineveryaspectofhumanlife.ProgrammingcomputerstounravelthishiddeninformationiswhatMachineLearningisallabout.Itistheartandscienceofscientificallyderivinginsights,patternsandpredictionsfromdata. Thoughithasbeenanareaofactiveresearchforover50years,MachineLearningiscurrentlyundergoingarenaissancedrivenbyMoore'slawandtheriseofbigdata.Largeprivateandpublicinvestmentintheareahasgivenusself-drivingcars,practicalspeechrecognition,effectivewebsearch, andavastlyimprovedunderstandingofthehumangenome.ComputerbasedMachineLearningalgorithmsnowoutperformhumansontaskssuchashandwrittendigitrecognition,trafficsignrecognition,andevenonsomecomplexreasoningtasksasdemonstratedbyIBM'sWatsonwinningJeopardy. MachineLearningmodelsandprogramsautomaticallymakedecisionsfromdatainordertoachievesomegoalorrequirement.Machinelearningmodelsmattertotheworld.Becausetheyare; #EFFICIENT MachineLearningmodelspredictanddetectpartnersfasterthananyothermanualprogramormethod. #EFFECTIVE MachineLearningmodelscandobetterjobthanhumanswhenanalysingandpredictinglargescaleandstreamingdatasets(bigdata). #SCALE MachineLearningmodelscanprovidesolutionstolargedataproblemsthattraditionalsystemscannotsolve.
- 4. www.persontyle.com © 2014 Persontyle Ltd. All rights reserved. Machine perception Computer vision, includingobject recognition Natural language processing Pattern recognition Search engines Medical diagnosis Bioinformatics Brain-machine interfaces Detectingcredit card fraud Stock marketanalysis ClassifyingDNA sequences Sentiment analysis Affective computing Information retrieval Recommender systems Examplesintherealworldincludehandwrittenrecognition, weatherprediction,frauddetection,search,facialrecognition,andsoforthareallexamplesofmachinelearninginthewild. ApplicationsforMachineLearninginclude: “OverthepasttwodecadesMachineLearninghasbecomeoneofthemainstaysofinformationtechnologyandwiththat,arathercentral,albeitusuallyhidden,partofourlife.Withtheeverincreasingamountsofdatabecomingavailablethereisgoodreasontobelievethatsmartdataanalysiswillbecomeevenmorepervasiveasanecessaryingredientfortechnologicalprogress.” DR. ALEXANDER J. SMOLA, PROFESSOR, CARNEGIE MELLON UNIVERSITY
- 5. www.persontyle.com © 2014 Persontyle Ltd. All rights reserved. MachineLearningenablescomputationalsystemstoadaptivelyimprovetheirperformancewithexperienceaccumulatedfromtheobserveddata. Thoughithasbeenanareaofactiveresearchforover50years,MachineLearningiscurrentlyundergoingarenaissancedrivenbyMoore'slawandtheriseofbigdata.Largeprivateandpublicinvestmentintheareahasgivenusselfdrivingcars,practicalspeechrecognition,effectivewebsearch,andavastlyimprovedunderstandingofthehumangenome. Computerbasedmachinelearningalgorithmsnowoutperformhumansontaskssuchashandwrittendigitrecognition,trafficsignrecognition,andevenonsomecomplexreasoningtasksasdemonstratedbyIBM'sWatsonwinningJeopardy. FundamentalsofMachineLearningBootcampwilltakeyouthroughtheconceptualandappliedfoundationsofthesubject.TopicscoveredwillincludeMachineLearningtheory,typesoflearning,techniques,modelsandmethods.LabsaredevelopedtopracticallylearnhowtousetheRprogramminglanguageandpackagesforapplyingthemainconceptsandtechniquesofMachineLearning. Overthecourseoffivedays,overtwodozentechniqueswillbeexamined, implementedthroughsupervisedexercisesandtutorials,andcompared. Youwilllearntherelativeadvantagesanddisadvantagesofdifferenttypesoftechniquesindifferentcontexts.Youwillseehowsomemodelsareentirelydatadriven,whileotherscanbeusedtoencodedefeasibleexpertknowledge.Youwilllearnmethodsforvalidatingselectedmodelsandtechniquesandforchoosingamongalternativemethods. FUNDAMENTALS OF MACHINE LEARNING BOOTCAMP
- 6. WHAT WILL YOU LEARN? In this bootcampyou will learn, among other things: +What Machine Learning entails and why it is important +The different types of Learning, especially Supervised Learning +Be able to use R to apply a number of the most common and powerful statistical machine learning techniques. +Know how to implement such techniques in principle and therefore be able to apply their knowledge within paradigms outside R. +Be able to appreciate the trade-offs involved in choosing particular techniques for particular problems. +Be able to utilize rigorous methods of model selection. +Understand the mathematical ideas behind, and relationships between, the various methods. +Have a greater confidence in their knowledge and standing as a data scientist. +How to use these algorithms in a variety of benchmark datasets +How to fine-tune these algorithms for better performance www.persontyle.com © 2014 Persontyle Ltd. All rights reserved. R logo is trademark of the R Foundation, from http://www.r-project.org PREREQUISITES KnowledgeofRprogramminglanguageandfamiliaritywithlinearalgebra. Basicfamiliaritywithstatisticsandprobabilitytheoryisrecommended.
- 7. SCHEDULE AND LEARNING OBJECTIVES www.persontyle.com © 2014 Persontyle Ltd. All rights reserved. Time Topic/Activity 09:00-09:30 Introduction 09:30-11:00 1. R Refresher 11:00-13:00 2. Linear and Quadratic Regression After this module, you will: •Understand what regression is. •Understand what linearity is. •Understand the idea behind basis projection. •Be able to perform linear, quadratic and polynomial regression. •Be able to identify datasets that are suitable for linear and quadratic regression. •Understand the idea of free parameters. 13:00-13:30 Lunch 13:30-15:00 2. Principle Component Analysis After this module, you will: •Understand how PCA functions. •Understand how PCA can be used for feature selection and information compression. •Be able to perform PCA analysis and regression. •Understand and be able to perform scaling and centring of data. 15:00 -15:15 Coffee Break 15:15-17:15 3. Feature Selection and Shrinkage After this module, you will: •Understand the idea of feature shrinkage •Be able to use subset selection as a means of feature selection •Be able to use Ridge Regression and the Lasso as means of feature shrinkage. •Understand what degrees of freedom are. •Understand what the variance/bias trade-off is. •Have a basic understanding of how both relate to the question of model selection. 17:15-18:00 4. Error Estimation After this module, you will: •Know what residuals are •Be able to model regression error using a normal distribution. DAY 1 DAY 2 DAY 3 DAY 4 DAY 5
- 8. SCHEDULE AND LEARNING OBJECTIVES www.persontyle.com © 2014 Persontyle Ltd. All rights reserved. Time Topic/Activity 9:00-11:00 5. Real-Discrete Classification: LDA, QDA and Logistic Regression After this module, you will: •Understand what classification tasks are, and the difference between real- discrete and discrete-discrete classification. •Be able to apply LDA, QDA and Logistic Regression. 11:00-11:15 CoffeeBreak 11:15-13:00 6. Perceptron Classification After this module you will: •Understand how to use the perceptron classifier in separable and inseparable cases. •Understand the idea of linearly separable and inseparable data. •Understand the idea of iterative algorithms and termination conditions. 13:00-13:30 Lunch 13:30-15:30 6. Discrete-Discrete Classification & An Introduction to Bayesian Methods After this module, you will: •Be able to apply conditional multinomial and noisy-or models to discrete- discrete classification tasks. •Understand the idea behind Bayesian Methods in statistics •Be able to work with Dirichletpriors, and understand the idea of count and pseudo-count parameters. 15:30-15:45 Coffee Break 15:45-17:45 7. K-Means and Cluster Analysis After this module, you will: •Understand and be able to compute the distance between data points. •Understand unsupervised learning and cluster analysis. •Be able to apply the K-Means and K-Mediodalgorithms for cluster analysis. DAY 1 DAY 2 DAY 3 DAY 4 DAY 5
- 9. SCHEDULE AND LEARNING OBJECTIVES www.persontyle.com © 2014 Persontyle Ltd. All rights reserved. DAY 1 DAY 2 DAY 3 DAY 4 DAY 5 Time Topic/Activity 9:00-11:00 8. K Nearest Neighbours After this module, you will: •Understand what is meant by local methods, their weakness regarding memory use, and the situations in which they are suitable •Be able to apply the K-Nearest-Neighbours and Adaptive K-Nearest-Neighbours techniques 11:00-11:15 Coffee Break 11:15-13:00 9. Local Regression After this module, you will: •Be able to perform local regression. 13:00-13:30 Lunch 13:30-15:30 10. Kernel Density Estimation After this module, you will: •Understand what a kernel is. •Be able to identify common kernels. •Understand what bandwidth is and why it is important. •Be able to perform kernel density estimation. •Understand what thinning is and be able to perform thinned kernel density estimation using K-Means or K-Mediods. •Be able to identify cases where kernel density estimation is suitable. 15:30-15:45 Coffee Break 15:45-18:00 11. Regression/Classification Trees and Boosting After this module, you will: •Understand and be able to implement regression/classification trees. •Understand what boosting is. •Be able to implement the adaboostalgorithm.
- 10. SCHEDULE AND LEARNING OBJECTIVES www.persontyle.com © 2014 Persontyle Ltd. All rights reserved. DAY 1 DAY 2 DAY 3 DAY 4 DAY 5 Time Topic/Activity 9:00-11:30 12-Splines After this module, you will: •Understand what truncated exponential splines are and how we can use bases projection to calculate them. •Understand the border issues associated with regression splines and how natural splines assist in dealing with these. •Understand what B-Splines are and how they are used. •Be able to use truncated exponential regression and natural splines, as well as B-Splines. •Be able to work with tensor products of such splines 11:30-13:00 13. MARS After this module, you will: •Be able to use the MARS procedure for working with splines. •Be able to identify cases where such additive methods are appropriate. •Understand the idea of effective degrees of freedom. 13:00-13:30 Lunch 13:30-14:15 AzureMachine Learning Studio Overview –1 14:15-16:30 14. Smoothing / Thin Plate Splines After this module, you will: •Understand what smoothing splines are, their optimality guarantees and their complexity issues. •Understand the connection between penalizing the second derivative of smoothing splines and performing Ridge Regression on a transform of the dataset. 16:30-18:30 15. Radial Basis Networks After this module, you will: •Understand what radial basis functions and networks are, how they make use of kernels to project our data to new bases and the connection with ridge regression to smooth the resulting models. •Be able to use Radial Basis Networks to model data. •Be able to use appropriate thinning strategies to avoid the complexity issues identified.
- 11. SCHEDULE AND LEARNING OBJECTIVES www.persontyle.com © 2014 Persontyle Ltd. All rights reserved. DAY 1 DAY 2 DAY 3 DAY 4 DAY 5 Time Topic/Activity 09:15-10:15 16. Support Vector Classifiers After this module, you will: •Know what support vectors, optimal hyperplanesand support vector classifiers are. 10:15-12:15 17. Support Vector Machines After this module, you will: •Understand how SVMs work, the reasons for their success, and the links between them and simpler statistical models from earlier modules. •Be able to apply support vector machines to appropriate cases. 12:15-13:00 AzureMachine Learning Studio Overview –2 13:00-13:30 Lunch Break 13:30-16:45 18. Neural Networks After this module, you will: •Understand how Neural Networks work, the reasons for their success, and the links between them and simpler statistical models from earlier modules. •Be able to train Neural Networks for classification and regression tasks using the back-propagation algorithm with weight decay. •Be able to apply Neural Networks to appropriate cases. 16:45-18:15 19. Model Selection After this module, you will: •Be able to apply validation and information criteria model selection methods to real life problems. •Understand the advantages and disadvantages of the different methods. •Understand the relationship between model fitness and complexity measures such as effective degrees of freedom.
- 12. www.persontyle.com © 2014 Persontyle Ltd. All rights reserved. Persontyletrainersarepassionateaboutmeetingeachparticipantslearningneeds.TheyhavebeenchosenbothfortheirextensivepracticalDataScienceandMachineLearningexperienceandfortheirabilitytoeducateandinteractwithnaturalempathy.AllofourtrainershaveworkedonavarietyofdatascienceandMachineLearningprojects.Theysharetheiracademicknowledgeandreal-worldexperienceandeachindividualaddstheirownuniqueperspectivetothecourse.Ourtrainerspresentinastylethatisinformal,entertainingandhighlyinteractive. GuestSpeakers Businessleaders,MachineLearningpractitioners,andacademicresearcherscoveringusecases,casestudiesandsharingpracticalexperienceofapplyingDataScienceandMachineLearningintheirorganizations. COURSE INSTRUCTORS “A BREAKTHROUGH IN MACHINE LEARNING WOULD BE WORTH TEN MICROSOFTS” BILL GATES, CHAIRMAN, MICROSOFT WHO SHOULD ATTEND AnyoneinterestedinlearningandapplyingmachinelearningmethodsandRtosolvereal-worlddataproblems.Idealforpeopleinterestedinpursuingcareerindatascience. Thishands-onworkshopisaimedatbusinessandtechnologyprofessionals,Developer,Architect,Manager,DataAnalyst,BIDeveloper/Architect,QA,PerformanceEngineers,Sales,PreSalesandMarketing,ProjectManager,PublicServices,TeachingStaffandallthosewhoalreadyhavesomebasiccompetenceinstatisticsbutwishtobeginusingRformachinelearningforthefirsttime.
- 13. For corporate bookings or to organize on-site training email hello@persontyle.comor call now +44 (0)20 3239 3141 Register Now RETURN ON INVESTMENT (ROI) CONVINCE YOUR BOSS The advent of the data driven connected era means that analyzingmassive scale, messy, noisy, and unstructured data is going to increasingly form part of everyone's work. The School of Data Science learning programs provide a unique investment opportunity that pay’s for itself many times over. "For the best return on your money, pour your purse into your head." World- class Instructors Benjamin Franklin Develop Practical Data Science Skills Real World Industry Use Cases Short Courses For Time Convenience Value For Money THE SCHOOL OF DATA SCIENCE The School of Data Science, a project of Persontyle, specializes in designing and delivering structured, relevant and practical learning experiences for all of us to understand data science in simple human terms. Follow us on Twitter @schooltds Like us on Facebook Get in touch! hello@personyyle.com Limited seats. We encourage you toregister as soon as you can. WWW.PERSONTYLE.COM/SCHOOL