SURVEY ON BRAIN – MACHINE INTERRELATIVE LEARNING
Professor Shraddha Ovale, Dept. of Computer engineering, PCCOE, Maharashtra, India
Abstract - With the rapid development from machine learning (ML) to deep learning (DL) there is closer connection between human brain machine working. The purpose of this article is to provide survey on learning based human machine. We know how machine do work like brain. But how brain is consuming every data and do work like a machine? More specifically, first we will see how we learn something and after a period we do them as an involuntary action. Second we will see how the mathematical equations will work to define brain machine functionality. Finally, we will briefly analyse how AI ML and brain are interdependent.
Key Words: Machine learning, deep learning, automation, perceptron.
1. INTRODUCTION
1 Brain Automation
Brain makes judgments and decisions quickly and automatically. It continuously makes predictions about futureevents.Itlearnsfrompastfailuresandeliminatethose consequencesinfuturedecisions.
Nowadays lots of youths try to learn how to flip a pen in handwithhelpoffingersandpalm.Therearelotsoftricksto dothatandsomeareveryhard.Withtimewelearnthose veryfluentlyandevenwhilereadingabookwecanflipapen infingerseasily.Howdoesourbrainactuallyknowtodothis involuntarily? Instartingofflearningthisourbrainslowly collects data of when and where we have to change our fingerandwhichfingerwhileflipping.Withlotsofsuccessful data,the brain triesto removefailuredata,andafter long term,thebraincatchesperfectprobabilityandconsequences toflipthatpenwithoutfailureandfluently.Suchlotsofsmall dataiscollectedinpacketswhichislikewhenyou’llchange the size of pen brain will take very little time to handle it comparativelyittookforthefirsttime.It’salldependenton some algorithm that is used by our brain to act the same functionondifferentparameters.Asimilarworkfunctionis usedinMLbyusingAI.
Fromthedayofbirth,ourBrainneuronsstarttocollectdata. Asforasimplemachine,thereisnoinbuiltdataispresentin such a way our brain is quite empty to learn things and functions.Withreachingnearlyageof4 5ourbrainmakes such complexity of neuron networks which increase consumptionofdata that’s whywestartourschool atthe ageof4 5yearsold.Attheageof18,wedonoteventhink that we are actually walking on stairs and we do it
involuntarilybutwhenwewerechildren,welearneditvery toughly.SimilarevolutionwecanseeinmachineswithAI.
1.1 Neural network in ML which mimics Brain:
Neural Network in ML is just a carbon copy of our Brain system.Ahumanbrainhavebillionsof neurons[5],theyare connectedtoeachother.Humanneuronsarecellssowhen onegetsactivateditsendsignalstootherinitsnetwork.
Likea human brain,the machinelearning neural network also consists of interconnected neurons. When a neuron receivesinputs, itgetsactivatedandsendsinformation to otherneurons.
Duetotheplasticityofourbrain,wedotasksandbecome better at them and such thing is Machine learning. For example,whenwelookatapictureofadog,weknowthatit is a cat because we have seen enough dogs in our lives. Likewise, if we provide our neural networks with enough dogimages[1],theywillstarttorecognizedogs.
1.2 AI and Neuroscience
Neuralnetworksactas“virtualbrains”[5]theydofunctions likeourbrain.Dataisfeedbygivinglotsofsimilarimagesto virtual machine to recognize some pattern which further helpsneuroscientiststodotheircalculationsandhypothesis.
However,thewayartificialintelligencesystemsworkisvery differentfromourbrain.Asourbrain[2]isabiologicalpart thatisverydifferentthanapieceofmachinery.Andusing patternrecognitionwithhelpofanAImachineworkslikea neuralnetworkofthebrain.
Arif Mulani, Abhishek Pawar, Sairaj Pawar, Bhushan Pisal
International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056
Volume: 09 Issue: 04
2022 www.irjet.net p ISSN: 2395 0072
2 Mathematics of artificial neural network:
Artificial neuron (Perceptron):
An artificial Neuron is not a physical object [8] but it is a mathematical structure that mimics the function of actual brainneuronstosomeextent.Itconsistsoflotsofnodesas realneuronshaveDendrites.
● Simpleworkofperceptron(anartificialneuron):
1)NodestakeNnumberofinputs.
2)Addallinputs.
3)Multiplyeachinputwithitsweight.
4)thenaddupalltheproductsofinputandweightswhich givesaweightedsum.
5)Ifthe weightedsumis greater[9]than some threshold value returns a decision of 1 and otherwise returns a decisionof0,similartoneuronfiringandnotfiring.
We’ve seen how perceptron takes decisions based on differentinputsbuthowactuallyitlearnsnewthings?How diditfindtherightparameters?
Ify(x∙w) 0: w=w+yx
y=Label(either 1or+1),x=currentdatapoint,(x∙w)=dot product(Weightvector).
Ifexpressiony(x∙w)islessthanorequaltozerothenlabely is different than the predicted label let say z(x ∙ w) so we updateweightsasw=w+yx.
How update rule works?
y(x∙wᵢ)=y(x∙(w+yx))
=y(x∙w+y||x||²) =y(x∙w)+y²||x||² =y(x∙w)+||x||²
As||x||²≥0=>y(x∙wᵢ)≥y(x∙w)
Due to square factors, new value will be near to positive valueandhenceitiscorrectlyclassified.
In the graphical representation, for 2 inputs decision boundaryisalinesimilarlyfor3inputsboundarywillbea2 D plane such that for N inputs (N 1) D decision boundary willexist.
Let’s see simplification by using 2 inputs:
(Workofperceptronon2inputs)
w’representsweightvectorwithoutbiasterm(Parameter used to represent patterns that do not pass through the origin.). This vector represents the slope of the decision boundary.
© 2022, IRJET
Factor
Drawback: PerceptronfailstoshowtheresultonXOR[3].Itcanonly classifylinearlyseparablesetsofvectors.Itjustgivesoutput valueslike0or1,TrueorFalse.Formorecomplexproblems werequiredmorethanoneperceptrontogetanoutput.
3. Interdependence between an AI and brain
The revolution of AI systems can help drive neuroscience forward and unlock the secrets of the brain. It allows researchers to build better models and theories of the human brain. Similarly, our recognition system helps machinestolearnthingsthroughAI.Probabilisticmodelling [7]alsohasadvantagesovertraditionalnormativetheoryin termsoflearninginAIsystem.
WiththehelpofAI,wecansolvelotsofproblemsthatweare not able to solve by simple thinking. As we have made calculatorssolvelongproblemsinshortterm.The“Towerof Hanoi”[10]isanexampleofthat.Actually,itisimpossibleto solvethisproblemwithlotsofdisks.Butwithhelpofdata structures,weareabletosolveit.
So,withhelpofAI,wearehelpingneurosciencetocrossthe boundaries to achieve lots of things and with our daily problems,weareenhancingthecapacityofAI.
9001:2008
Literature Review
1.MathematicsofNeutralNetworkpaperbyGabrielPeyréet al [3] explains the algorithm and mathematics behind the learningprocess.Inordertoexplainthis,twotypesofneural networks are mentioned and considered here i.e 1]Discriminative Neural Network, 2]Generative Neural Network.InDiscriminativeNeuralNetworkparametersin terms of dimensions are set to define an object and then latertoidentifyanunknownobject,theparameterofthat object are considered and probability is calculated using mathematicalexpressions,basedonthisprobabilityofthe unknownobjectisidentified.InthecaseoftheGenerative NeuralNetworkparameterintermsofcolorandfeatureand further using this parameter the unknown object is identified.
2.BrainOS:ANovelArtificialBrain AlikeAutomaticMachine Learning Framework paper by Newton Howard et al [5] explainsthehypothesisregardingcreatingartificialhuman intelligenceinamachinelearningframework.TheBrainOS systemmimicsthehumanthinkingprocesswhichismuch differentthantheotherMLalgorithmsandtools.Duetoits abilitytothinklikehumans,itcansolvehigh levelproblems.
3.CommonlyusedMachineLearningAlgorithmanarticleby Sunil explainssomeofthealgorithmsofmachinelearning and also explains it works. Some of these algorithms are linear regression, logistic regression, decision tree, SVM, Naïve Bayes, k means, random forest, dimensionality reductionalgorithms,gradientboostingalgorithms.
4. Mathematics of artificial neutral networks article on Wikipediaexplainstheartificialneuralnetwork’sprinciple, structure, algorithm, and working. It also explains neural networksinmathematicalexpressionbymakingafunction ofthem.
5.Fascinating Relationship between AI and Neuroscience articlebyHong Jing explainsus theroleandcontribution playedbyneuroscienceinAIdevelopmentbygivingsome real lifeexamples,butalongwithitexplainshowAhelpsto understandhowthehumanbrainworks
6.The Brain’s Autopilot Mechanism Steers Consciousness articlebySteveAyantalksaboutthebrain’sabilitytotakea decision unconsciously and its mechanism behind this unconsciousdecisionmakingpowerandlearningofvarious actionsandperformingtheminvoluntarily,isexplainedby referringtovarioustheoriesinofneurobiology,oneofthem isthetheoryofpredictivemind.Accordingtothetheoryof predictivemind,therearevariousautomaticprocessesand thoughts going on in the brain and with the help of this involuntaryprocess,thebrainisabletopredictasituation andeventquicklyandaccurately.
7.Probabilistic machine learning and artificial intelligence paper by Zoubin Ghahramani depicts the importance of
probabilistic programming in the fields of Artificial Intelligence and Machine Learning. Probabilistic programmingusesthetheoryofprobabilityandprobability distributioneithertoidentifyapattern,object,structure,etc ortoextractunknowninformationaboutitortocarryout thetaskassignedtoit,onthebasisofchancesofthefeature, shape,size(inordertoidentifyimage),orthetypeofthe condition given to it (in order to analyze data and find its applicationortosolveproblemsandbugs).Aprobabilistic approachiswaymoreefficientthanthatofthetraditional normativetheory,astheprobabilisticapproachiswaymore flexibleandalongwiththat itisalsocapableofproducing datafromanymodelorsubjectirrespectiveofitsdomain.
8. Mathsinaminute:ArtificialNeuronarticlebyMarianne givesusinformationregardingthestatisticaldataregarding neuron. Byunderstandingthemathematics,structureand mechanismbehindthefunctioningofthebrain.Andbyusing this understanding, the concept of artificial neuron is introducedhere.
9. Perceptron: Explanation, Implementation and a Visual Example article by Dorian Lazar explain the concept of perceptron. Perceptron is the term which refer to the artificial neural network. It isanalgorithm whichactually mimicstheprocessofthebiologicalneuroni.e.totakeinput, computetheweightofthesumandthenpassitthrougha thresholdfunction,finallydisplayingresultasaoutput.Now on the basis of this result the machine can plot/place a decision boundary. Along with the mechanism and the mathematicsofthisalgorithm,writerhasalsoexplainedits application and implementation of it along with some examples.
International
ID Related work Algorithm Result Advantages Limitations
[1] Invention of machinelearning Supervised learning
It starts to give idea about machine doing work itself without a handler.
[2] Mathematics of artificial neural network
[3] Commonly used machine learning algorithms
ANN The proposed model shows mathematical proposition of an artificialneuron.
A new way of making mechanical works, calculation, and various activities with artificialbrain.
There are lots of difficulties about algorithms, structure, price, processing capacity.
The artificial neural networklearnsbyitself and can store information within it insteadofondatabase.
Wecan’tclaimthatitis more efficient than statisticalmethod.
Linear regression, SVM, KNN, etc.
Invariousapps,software, games machine learning uses algorithm based on choices, probability which improve itself by gathering the data and making further process welloriented.
Wedon’thavetospend any more time after making algorithm full fledgeditcanimprove itself.
Asitismakingchoices byitself,wehaveslight chance of error due to itslackofknowledgeor lessdataavailable.
[4] AIandneuroscience, building neural network that will mimicbrain.
NLP
Wemadeneuralnetwork that will resemble to brain. And can make decisions and learn throughit.
Googleabletobuildan AIequivalenttoa55.5 IQhuman.
Itcannotmatchbrain’s high dimensional matrix.Scientisttrying to understand how billions of neurons work together and makes a complex structure.
[5] Unconscious decisions Predictive mind There are some automatic processes which leads the mind to thespecificdecision.
[6] Probabilistic machinelearning HMM Probabilistic programming uses theoryofprobabilityand probabilitydistribution.
[7] Perceptron perceptron
Mathematical representation of artificialneuron
Mechanism like this willbeabletopredicta situation as react so quicklyandaccurately.
More efficient and flexible approach. Producing data from anymode;irrespective ofitsdomain.
nonlinear problems and complex patterns in data are possible to process.
Uncontrolled processes.
Non probabilistic problems are required differentapproach.
Still not efficient to process thousands of inputs and 10 degree polynomials.
International
3. CONCLUSIONS
MachineAutomationplaysanimportantroleintheglobal economy and daily lifestyle. Scientists trying to modify automated devices with mathematical tools to create complex neural systems for a rapidly expanding range of applications and human activities. This survey tries to explain the mathematics and the mechanism behind the machinelearninganditsrelationwithbrain.Theupcoming technologyhaspotentialtoprocesshighdimensionaldata andthousandsofinputs.
REFERENCES
1.GabrielPeyréCNRSetal.MathematicsofNeuralNetworks.
2.NewtonHoward,NaimaChouikhietal.BrainOS:ANovel Artificial Brain Alike Automatic Machine Learning Framework. Front.Comput.Neurosci.14:16. doi: 10.3389/fncom.2020.00016
3. URL:https://www.analyticsvidhya.com/blog/2017/09/com mon machine learning algorithms/
4.Mathematicsofartificialneuralnetworks.Artificialneural network Wikipedia.
5.Jingles(HongJing)FascinatingRelationshipbetween AI andNeuroscience.PublishedinTowardsDataScience Mar3,2020.
6.TheBrain’sAutopilotMechanismSteersConsciousnessBy Steve Ayan on December 19, 2018. URL: https://www.scientificamerican.com/article/the brains autopilot mechanism steers consciousness/
7.Probabilisticmachinelearningandartificialintelligence by Zoubin Ghahramani, University of Cambridge, May 28, 2015.
8. URL: https://plus.maths.org/content/maths minute artificial neurons
9. Perceptron: Explanation, Implementation and a Visual Example by Dorian Lazar on Apr 6, 2020URL:https://towardsdatascience.com/perceptron explanation implementation and a visual example 3c8e76b4e2d1
10.URL:https://en.wikipedia.org/wiki/Tower_of_Hanoi
BIOGRAPHIES:
Arif Mulani, FY B.tech (Comp. Engineering),PCCOE.
AbhishekPawar,FYB.tech(Comp. Engineering),PCCOE.
Sairaj Pawar, FY B.tech (Comp. Engineering),PCCOE.
Bhushan Pisal, FY B.tech (Comp. Engineering),PCCOE.
