deep learning evaluation and its advantages.ppt

deep learning

1. 1
DEEP LEARNING

2. CONTENTS
I.Introduction
II.History
III.Principle
IV.Technology
V.Working
VI.RealTimeApplications
VII.FutureScope
VIII.Advantages
IX.Conclusion
2

3. INTRODUCTION
WhatisDeepLearning?
Deeplearningisabranchof
machinelearningthatuses
data,loadsandloadsofdata,to
teachcomputershowtodo
thingsonlyhumanswere
capableofbefore.
Forexample,howdomachines
solvetheproblemsof
perception?

4. HISTORY
4
1958:Frank Rosenblatt
creates the perceptron, an
algorithm for pattern
recognition.
1989:Scientists were able to
create algorithms that used deep
neural networks.
2000's:The term “deep
learning” begins to gain
popularity after a paper by
Geoffrey Hinton.
2012:Artificial pattern-
recognition algorithms achieve
human-level performance on
certain tasks.

5. PRINCIPLE
Deeplearningisbasedontheconceptofartificialneural
networks,orcomputationalsystemsthatmimicthewaythe
humanbrainfunctions.

6. TECHNOLOGY
Deeplearningisafast-growingfield,andnewarchitectures,
variantsappeareveryfewweeks.We'llseediscussthemajor
three:
1.ConvolutionNeuralNetwork(CNN)
CNNsexploitspatially-local
correlationbyenforcingalocal
connectivitypatternbetween
neuronsofadjacentlayers.

7. TECHNOLOGY
2.RecurrentNeuralNetwork(RNN)
RNNsarecalledrecurrentbecausetheyperformthesametask
foreveryelementofasequence,withtheoutputbeing
dependedonthepreviouscomputations.OrRNNshavea
“memory”whichcapturesinformationaboutwhathasbeen
calculatedsofar.

8. TECHNOLOGY
3.Long-ShortTermMemory
LSTMcanlearn"VeryDeepLearning"tasksthatrequire
memoriesofeventsthathappenedthousandsorevenmillionsof
discretetimestepsago.LSTMworksevenwhentherearelong
delays,anditcanhandlesignalsthathaveamixoflowandhigh
frequencycomponents.
8

9. WORKING
Considerthefollowinghandwrittensequence:
Mostpeopleeffortlesslyrecognizethosedigitsas504192.That
easeisdeceptive.
Thedifficultyofvisualpatternrecognitionbecomesapparentif
youattempttowriteacomputerprogramtorecognizedigitslike
thoseabove.
9

10. WORKING

11. WORKING
Theideaofneuralnetworkis
todevelopasystemwhichcan
learnfromtheselargetraining
examples.
Eachneuronassignsa
weightingtoitsinput—how
correctorincorrectitisrelative
tothetaskbeingperformed.
Thefinaloutputisthen
determinedbythetotalof
thoseweightings
A training
Sample
A very basic approach:
Binary Classifier

12. REAL TIME APPLICATIONS
AutomaticColorizationofBlackandWhiteImages
AutomaticallyAddingSoundsToSilentMovies
AutomaticHandwritingGeneration

13. REAL TIME APPLICATIONS
AutomaticTextGeneration
AutomaticImageCaptionGeneration
AutomaticGamePlaying

14. FUTURE SCOPE
1.DeepLearningwillspeedsearchforextraterrestriallife.
RobERt,shortforRoboticExoplanetRecognitionforExoplanets
thatarebeyondoursolarsystem.

15. FUTURE SCOPE
2.ForAstronauts,NextStepsonJourneytoSpaceWillBe
Virtual
3.DroughtsandDeepLearning:MeasuringWaterWhereIt’s
Scarce

16. ADVANTAGES
1.Itdoesfeatureextraction,noneedforengineeringfeatures
2.Movingtowardsrawfeatures
3.Betteroptimization
4.Anewlevelofnoiserobustness
5.Multi-taskandtransferlearning
6.BetterArchitectures

17. CONCLUSION
ThelowmaturityofDeepLearningandits
applicationssuchaslargedeepneural
networksachievethebestresultsonspeech
recognition,visualobjectrecognitionand
severallanguagerelatedtaskfieldwarrants
extensivefutureresearch.Nevertheless,the
possibilitiesofdeeplearninginfutureare
infiniterangingfromdriverlesscars,torobots
exploringtheuniverseandtowhatnotifthe
upcomingarchitecturesarecreativeenough.

18. 18
THE
END….