as you know we are living in a world ofhumans and machine humans have beenevolving and learning from the pastexperience for millions of years on theother hand the era of machine and robotshave just begun now you can consider itin a way that currently we are living inthe primitive age of machines while thefuture of machines is enormous and isbeyond a scope of imagination now intoday’s world these machine or therobots have to be programmed before theystart following your instructions butwhat if the Machine started learning ontheir own from their experience worklike us feel like us do things moreaccurately than us might even start awar of their own now these things soundsfascinating and a little scary rightlet’s just remember this is just thebeginning of the new era now let’ssuppose one day you went for shoppingmangoes the vendor had a cart full ofmangoes from where you could handpickthem weigh them and pay them accordingto the fixed rate now the questionarises is how will you choose the bestmangoes you were informed that brightand yellow mangoes are sweeter than peeland the yellow ones so you make a simplerule pick only from the bright yellowmangoes you check the color of themangoes pick the bright yellow ones payup and return home right now when youwent home and tasted all the mangoessome of them are not sweet as youthought you concluded that when it comesto shopping them you have to look formore than just the colors after a lot ofpondering and tasting different types ofmangoes you concluded that the biggerand brighter yellow mangoes areguaranteed to be sweet while the smallerbright yellow mangoes are sweet onlyhalf the time the next time at themarket you see that your favorite vendorhas gone out of town now you decide tobuy from a different vendor who suppliesmangoes born from a different part ofthe country now you realize that therule which you had learned that the bigred yellow mangoes are the sweetest isno longer applicable here you madeanother observation here that at thisparticular vendor that soft mangoes arethe juiciest now let’s suppose you goout with your girlfriendand she does not even like mangoes youknow how girlfriends are right and shewould like you to buy oranges for hernow all your accumulated knowledge ofwhat man was is worthless at this pointof time now you have to learn everythingabout the correlation between thephysical characteristic and the taste ofthe oranges by the same method ofexperimentation but then again this isnot as difficult as you thought but whatif you have to write a quote for it soas humans you would write a chordsomething like this if mango is brightyellow and the size is big that impliesthe mango is sweet and if the mango issoft that implies the mango is juicy nowconclusion as a human is that every timeyou make a new observation from yourexperiments you have to modify the listof rules manually you have to understandthe details of all the factors affectingthe quality of the mangoes if theproblem gets complicated enough it mightget difficult for you to make accuraterules by handthat covers all the possible types ofmangoes now this will take a lot ofresearch and effort and not everyone hasthis amount of time so this is wheremachine learning comes into picturewell machine learning is a concept whichallows the machine to learn fromexamples and experience and that toowithout being explicitly programmed soinstead of you writing the code what youdo is feed the data to the genericalgorithm and the algorithm or themachine will still logic based on thegiven data now let’s have a look at someof the features of machine learningwhich makes our life much more easier sowhat it does is that it uses the data todetect patterns in a data set and addjust the program action accordingly itfocuses on the development of thecomputer programs that can teachthemselves to grow and change whenexposed to new data it enables computerto find hidden insights using iterativealgorithm without being explicitlyprogram so know machine learning playsan important role in our day-to-day lifeas well you might not know it but youare surrounded by a lot of examples ofmachine learning and a lot of which issomething that you cannot live withoutfor example the first one is Google Mapsnow Google Maps is probably the app weuse whenever you go out and requireassistancein the direction and traffic now theother day I was traveling to anothercity and took the expressway and the mapsuggested despite the heavy traffic youare on the fastest route but that wasfine for me but how does it know thatwell it’s a combination of peoplecurrently using the service the historicdata of the route collected over thetime and few tricks acquired from othercompanies now everyone using maps isproviding their location the averagespeed the route in which they aretraveling which in turn has Googlecollect massive data about the trafficwhich makes them predict the upcomingtraffic and adjusts your route accordingto it now another application is theproduct recommendation but suppose youcheck an item on Amazon but you do notbuy it then and there but the next dayyou are watching videos on YouTube andsuddenly you see an ad for the same itemyou switch to Facebook chatting withyour friends and there also you see thesame ad so how does this happen wellthis happens because Google tracks yoursearch history I recommends ads based onyour search history this is one of thecoolest application of machine learningin fact you won’t believe that 35% ofAmazon’s revenue is generated just onlyby product recommendation well here isone of the coolest application ofmachine learning by far it is here andpeople are already using it and that isthe self-driving cars now machinelearning plays an important role in theself-driving car and I am sure you guysmight have heard about Tesla the leaderin this business and their currentartificial intelligence is driven by thehardware manufacturer in media which isbased on a type of machine learningwhich is the unsupervised learningalgorithm now there are certain stepswhich any machine learning algorithm hasto follow so the first step is datacollection and this stage involves thecollection of all the relevant data fromvarious sources now the second stepafter collecting all the data is datawrangling which is the process ofcleaning and converting the raw datainto a format that allows convenientconsumption now after the data have beencleaned and converted into a particularformat the data is analyzed to selectand filter the data required to preparethem all because not all the data isrequired for a particular model you haveto select certain featuresnow after selecting the features thealgorithm is strained on the trainingdataset through which the algorithmunderstands the pattern and the ruleswhich govern the data after this thetesting dataset determines the accuracyof our model and after this F model isready so the final stage comes is thatthe speed and the accuracy of the modelis acceptable then that model should bedeployed in the real system and afterthe model is deployed based upon itsperformancethe model is updated and improved and ifthere’s a dip in the performance themodel is retrained the machine learningis broadly classified into three majortasks which are the supervised and superand the reinforcement learning thesimplest from a machine learning is thesupervised learning and it is the onewhere you have input variables like Xand an output variable Y you use analgorithm to learn the mapping functionfrom the input to the output so insimple terms it implies y equals f of Xnow the goal is to approximate themapping functions so well that wheneveryou get some new input data X themachine can easily predict the outputvariables Y for the data now let merephrase this in simple terms insupervised machine learning algorithmevery instance of the training data setconsists of input attributes andexpected outputs the training data setcan take any kind of data as input likevalues of datasets rows the pixel of animage or even audio frequency histogramnow let me tell you why this category ofmachine learning is termed as supervisedlearning now this is because the processof an algorithm learning from thetraining data set can be thought of asthe teacher teaching his students thealgorithm continuously predicts theresult on the basis of the training dataand is continuously corrected by theteacher the learning continues until thealgorithm is an acceptable level ofperformance now any speech recognitionor any speech automated system on yourmobile phonetrains your voice and then startsworking based on this training data thisis an application of supervised learningbiometric attendance you can train themachine with inputs of your biometricidentity it can be your thumb your irisor your face for the matter of fat oncethe machine is trained it can validateyour future input and can easilyidentify you nowadays this is beingimplemented in all these smart phonesthat we have but sometimes the commanddata is unstructured and unlabeled so itbecomes very difficult to classify thatdata into different categories sounsupervised learning helps to solvethis problem now this learning is usedto cluster the input data into classeson the basis of the statisticalproperties now the training data is acollection of information without anylabel here now mathematicallyunsupervised learning is where you onlyhave the input data which is the X andno corresponding output variables nowthe goal of the unsupervised learning isto model the underlying structure or thedistribution in the data in order tolearn more about the data so we cameacross in a bottom point here which isclustering so what exactly is clusteringso clustering models focus onidentifying groups of similar recordsand labeling the records according tothe group to which they belong and thisis done without the benefit of priorknowledge about the groups and theircharacteristics in fact we may not evenknow exactly how many groups to look forbut the models are often referred to asunsupervised learning model since thereis no external standard by which tojudge the models classificationperformance there are no right or wronganswers to these models now marketbasket analysis is one of the keytechniques used by large retailers touncover association between items and itworks all on unsupervised learningit works by looking for combination ofitems that occurred together frequent inthe transaction not to put in anotherway it allows retailers to identify therelationships between the items thatpeople buy for example people who buybread also tend to buy butter now themarketing teams at the retail storesshould target customers who buy breadand butter and provide an offer to themso that they buy the third item like anegg so if a customer buys bread andbutter and sees a discount on or anoffer on egg he will be encouraged tospend more money and buy the eggsthis is what market basket analysis isall about the reinforcement learning isa part of machine learning where andagent is put in an environment and helearns to behave in this environment byperforming certain actions and observingthe rewards which it gets from thoseactions this reinforcement learning isall about taking appropriate action inorder to maximize the reward in aparticular situation in supervisedlearning the training data comprises ofthe input and so the model is trainedwith the expected output itself but whenit comes to reinforcement learning thereis no expected output the reinforcementagent decides what action to take inorder to perform a given task in theabsence of a training dataset it isbound to learn from its own experiencenow let’s understand this reinforcementlearning with an analogy to consider ascenario where and our baby is learninghow to walk now this scenarios can go intwo ways the first is that the babystarts walking and makes it to the candysince the candy is the end hold the babyis happy it’s positive reward now comingto the second scenario the waving startswalking but fails due to some harder inbetween the baby gets hurt even and doesnot get to the candy it’s negative thebaby is sad that you implies negativereward just like how we humans learnfrom our mistakes by trial and errorreinforcement learning is also similarwe have an agent which is here the babyand we have a reward which is the candywith many hurdles in between the agentis supposed to find the best possiblepath to reach the reward now anotherapplication of reinforcement learning isalso the games it is used to solve thedifferent games and sometimes achievesuperhuman performance but the mostfamous one must be the alpha go and theAlpha was zero it trained from thescratch and a researcher led the newagent alpha was zero play with itselfand finally beat the alpha go 100 tozero now this was a major breakthroughin the reinforcement learning processand also helped a lot of people in thedeep learning process as well and alsothe data scientist to make new robotsand create the artificial parts whichare there in the games so guys this wecome to an end of the sessionI hope you understood the basics ofmachine learning what it is what are thebasic types of machine learning how itis difficult for us to perform all ofthese scenarios by hand and write analgorithm by ourselvesso guys if you have any queriesregarding this session please feel freeto mention it in the comment sectionbelowtill then thank you and happy learning Ihope you have enjoyed listening to thisvideo please be kind enough to like itand you can comment any of your doubtsand queries and we will reply them atthe earliest do look out for more videosin our playlist and subscribe to anyrekha channel to learn more happylearning
Be Inspired By Data
