EE5M16MachineLearningwithApplicationsinMediaEngineering[5credits]Lecturer(s): UssherAssistantProfessorFrançoisPitié ModuleorganisationSemester Start

WeekEndWeek

AssociatedPracticalHours

Lectures TutorialsPerWeek Total PerWeek Total

1 1 12 20 2 22 1 11TotalcontactHours:53

ModuledescriptionThis module is an introduction to Machine Learning (ML), with a focus on DeepLearning.Deep learning isanothernameforartificialneuralnetworks,whicharealooselyinspiredbythestructureoftheneuronsinthecerebralcortex.AlthoughDeepLearninghasbeenaroundforquiteawhile,ithasrecentlybecomeadisruptive technology that has been unexpectedly taking over operations oftechnologycompaniesaroundtheworldanddisruptingallaspectsofsociety.WhenyoureadorhearaboutAIormachineLearningsuccessesinthenews,itreallymeansDeepLearningsuccesses.The course starts with an introduction to some essential aspects of MachineLearning,includingLeastSquares,LogisticRegressionandaquickoverviewofsomepopularclassificationtechniques.Then the course will dive into the fundamentals of Neural Nets, including FeedForwardNeuralNets,ConvolutionNeuralNetsandRecurrentNeuralNets.The material is constructed in collaboration with leading industrial practitionersincludingGoogle,YouTubeandMovidius,andstudentswillhaveguestlecturesfromthesecompanies.HandsonlabswillgiveyouexperiencewiththefieldandallowyoutodevelopyourownDeepLearningapplications.LearningoutcomesOncompletionofthismodule,thestudentwillbeableto:

1. Describe the main neural network architectures and parameters used inpopularDeepLearningsoftwarelibrariessuchasKeras.

2. Implementneuralnetworkapplicationsusingpython3andkeras

3. Evaluate the performance ofMachine Learning algorithms and analyse thepotentialpitfalls.

TeachingstrategyThe teaching strategy for this module is a mixture of lectures, problem-solvingtutorials and laboratories dedicated to implement and solve machine learningproblems.MostofthetheoreticalelementsofMachineLearningsandDeepLearningwill be covered in the first half of the term. The rest of the term is dedicated toexposuretomoreadvancedlabsandexposuretoindustryrelatedproblems.ProgrammingEnvironmentWehavedevelopedafantasticlabenvironmentspeciallyforthismodule,sothatyoucanlearnbestindustrypractices:

• programming will be done in python 3 using Keras and TensorFlow.Everything will be running on the Google Cloud Platform, which gives on-demandscalablecomputingresources.

• thecodingenvironmentwillbeacombinationofshell/terminal,editor,andJupyternotebook.

• Gitwillbeusedtocheckpointlabprogressandgivecontinuousfeed-backonlabassignments.

Wehavedevelopedaplatformthatwill smoothoutall thepainful installationandconfigurationpartssothatyoucanfocusontheessential.AssessmentThe finalwritten end-of-year examination counts for 75%of overallmark, the labsubmissionsfor20%andamid-termquizinweek8fortheremaining5%.Notethatit is expected that the student shouldbe able to complete theproblemsposed inlaboratoriesduringthetimetabledhours.Textbooks

• DeepLearning,IanGoodfellowetal.,(MITpress),[https://www.deeplearningbook.org]

• MachineLearningonCousera,AndrewNg[https://www.coursera.org/learn/machine-learning]

• NeuralNetworksandDeepLearning,MichaelNielsen[http://neuralnetworksanddeeplearning.com/]