Overview of Machine Learning Methods Prasanna Balaprakash Mathema5cs and Computer Science Division & Leadership Compu5ng Facility Argonne Na5onal Laboratory Argonne Training Program on Extreme-Scale Compu5ng (ATPESC) Aug 5th, 2017
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
OverviewofMachineLearningMethods
PrasannaBalaprakashMathema5csandComputerScienceDivision&
LeadershipCompu5ngFacilityArgonneNa5onalLaboratory
ArgonneTrainingProgramonExtreme-ScaleCompu5ng(ATPESC)Aug5th,2017
Acknowledgments
hPp://adilmoujahid.com/posts/2016/06/introduc5on-deep-learning-python-caffe/hPps://en.fabernovel.com/insights/tech-en/ai-for-dummieshPps://sebas5anraschka.com/blog/2016/model-evalua5on-selec5on-part2.htmlhPp://scoP.fortmann-roe.com/docs/BiasVariance.htmlhPp://videolectures.net/deeplearning2015_vincent_machine_learning/
2
Bornfromtheambi5ousgoalofar5ficialintelligence• Twohistoricallyopposedapproaches
Machinelearning
3
Traditional Programming
DataProgram
Output
AbundantData
OutputProgram
Givescomputerstheabilitytolearnwithoutbeingexplicitlyprogrammed
MachineLearning
Machinelearning
4
Ar7ficialIntelligence
MachineLearning
DeepLearning
DeepNeuralNetworks
Machinelearning
5
Magic, no, more like gardening • Seeds = Algorithms • Nutrients = Data • Gardener = You • Plants = Programs
InML,dataasalistofexamples(orturnitintoone)• ideallymanyexamples• preferablywitheachexampleavector(orfirstturnitintoone!)
Machinelearning
6
7
Machinelearning
Supervisedlearning
8
Classifica5on
9
10
Classifica5on
Regression
11
0 20 40 60
0 20 40 60 X
Y
?
11
Supervisedlearning:Genericframework
• 3elements– func7onfamily– lossfunc7on
• measurehowwronglythemodelpredicts
– searchforthebestfunc7on• mathema7calop7miza7on
12
Biasvariancetradeoff
13
• Allsupervisedlearningalgorithmsseektoreducebiasandvarianceinadifferentway
• Nofreelunch:nosinglealgorithmwillworkwellonalldataset
14
Machinelearning
Clustering
15
• Noexplicitpredic5ontarget• Usetheinherentstructuresinthedatatobestorganizethedataintogroups
ofmaximumcommonality(e.g.k-Means)
Dimensionreduc5on
16
• Noexplicitpredic7ontarget• Exploittheinherentstructureinthedatatosummarizeordescribedata
usinglessinforma7on(e.g.PrincipleComponentAnalysis)
Genera5veAdversarialNetworks
17
Genera5veAdversarialNetworks
18
19
Machinelearning
Reinforcementlearning
20
21
Machinelearning
ThankYou
22
www.mcs.anl.gov/~pbalapra
Related Documents
