Hello, everyone. This is Atul from Edureka and welcome to today's topicof discussion on AI vs Machine Learningvs Deep Learning. These are the termwhich have confused a lot of people and if youtoo are one among them, let me resolve it for you. Well artificial intelligenceis a broader umbrella under which machine learning and deep learning come youcan also see in the diagram that even deep learning isa subset of machine learning so you can say that all three of themthe AI the machine learning and deep learning are justthe subset of each other.
So let's move on and understand how exactly the differfrom each other. So let's startwith artificial intelligence. The term artificial intelligence was first coinedin the year 1956. The concept is pretty old, but it has gainedits popularity recently. But why well, the reason is earlier we hadvery small amount of data the data we had Was not enoughto predict the accurate result, but now there's a tremendousincrease in the amount of data statistics suggest that by 2020 the accumulatedvolume of data will increase from 4.4 zettabyte stewroughly around 44 zettabytes or 44 trillion GBs of data along with suchenormous amount of data. Now, we have moreadvanced algorithm and high-end computingpower and storage that can deal with such largeamount of data as a result. It is expected that 70% of Enterprisewill Implement ai over the next 12 months which is up from 40 percentin 2016 and 51 percent in 2017. Just for your understandingwhat does AI well, it's nothing but a technique that enables the machine to actlike humans by replicating the behavior and naturewith AI it is possible for machine to learnfrom the experience. The machines are justthe responses based on new input thereby performing human-like tasks. Artificial intelligence canbe trained to accomplish specific tasks by processinglarge amount of data and recognizing pattern in them. You can consider that building an artificialintelligence is like Building a Church, the first churchtook generations to finish. So most of the workers were working in it never sawthe final outcome those working on it took pridein their craft building bricks and chiseling stone that was going to be placedinto the great structure. So as AI researchers, we should think of ourselvesas humble brick makers whose job is to study how to build componentsexample Parts is planners or learning algorithmor accept anything that someday someoneand somewhere will integrate into the intelligent systemssome of the examples of artificial intelligencefrom our day-to-day life our Apple series just playingcomputer Tesla self-driving car and many more these examplesare based on deep learning and natural language processing. Well, this was about what is AIand how it gains its hype. So moving on ahead. Let's discuss about machinelearning and see what it is and white pros of an introduced. Well Machine learning came into existence in the late 80sand the early 90s, but what were the issueswith the people which made the machine learningcome into existence? Let us discuss them one by onein the field of Statistics. The problem was how to efficiently trainlarge complex model in the field of computer scienceand artificial intelligence. The problem was how to trainmore robust version of AI system while in the case of Neuroscience problemfaced by the researchers was how to design operationmodel of the brain. So these are some of the issues which had the largest influenceand led to the existence of the machine learning.
Now this machine learningshifted its focus from the symbolic approaches. It had inheritedfrom the AI and move towards the methods and model. It had borrowed from statisticsand probability Theory. So let's proceed and see what exactly ismachine learning. Well Machine learningis a subset of AI which The computer to act and make data-driven decisionsto carry out a certain task. These programs are algorithmsare designed in a way that they can learnand improve over time when exposed to new data. Let's see an exampleof machine learning. Let's say you wantto create a system which tells the expected weightof a person based on its side.
The first thing you dois you collect the data. Let's see there is how your data lookslike now each point on the graph representone data point to start with we can draw a simple line to predict the weightbased on the height. For example, a simple line W equal x minus hundredwhere W is waiting kgs and edges hide and centimeter this line canhelp us to make the prediction. Our main goal isto reduce the difference between the estimated valueand the actual value. So in order to achieve it wetry to draw a straight line that fits through allthese different points and minimize the error. So our main goal isto minimize the error and make them as small aspossible decreasing the error or the differencebetween In the actual value and estimated valueincreases the performance of the model furtheron the more data points. We collect the better. Our model will become we can also improve our modelby adding more variables and creating differentproduction lines for them. Once the line is created. So from the next timeif we feed a new data, for example heightof a person to the model, it would easily predict the datafor you and it will tell you what has predictedweight could be.
I hope you gota clear understanding of machine learning. So moving on ahead. Let's learn about deep learning. Now what is deep learning? You can consider deep learningmodel as a rocket engine and its fuel isits huge amount of data that we feed tothese algorithms the concept of deep learning is not new, but recently it's hype as increase and deep learningis getting more attention. This field is a particular kindof machine learning that is inspired by the functionality ofour brain cells called neurons which led to the conceptof artificial neural network. It simply takes the data connection between allthe artificial neurons and adjust them accordingto the data pattern more neurons are added at the size of the data is largeit automatically features learning at multiplelevels of abstraction. Thereby allowing a system to learn complex functionmapping without depending on any specific algorithm. You know, what no one actuallyknows what happens inside a neural networkand why it works so well, so currently you can callit as a black box. Let us discuss someof the example of deep learning and understand itin a better way. Let me start with a simpleexample and explain you how things happenat a conceptual level. Let us try and understand how you recognize a squarefrom other shapes. The first thingyou do is you check whether there are four linesassociated with a figure or not simple concept, right? If yes, we further check if they are connectedand closed again a few years. We finally check whether it is perpendicularand all its sides are equal, correct, if Fulfills. Yes, it is a square. Well, it is nothing buta nested hierarchy of Concepts what we did here wetook a complex task of identifying a square and this case and brokeninto simpler tasks. Now this deep learningalso does the same thing but at a larger scale, let's take an exampleof machine which recognizes the animal the taskof the machine is to recognize whether the given image isof a cat or a dog. What if we were asked to resolvethe same issue using the concept of machine learningwhat we would do first. We would Definethe features such as check whether the animal haswhiskers are not a check if the animal has pointed ears or not or whether its tailis straight or curved in short. We will Definethe facial features and let the system identify whichfeatures are more important in classifying aparticular animal now when it comes to deep learningit takes this to one step ahead deep learning automaticallyfinds out the feature which are most importantfor classification compare into machine learning where we Had to manually giveout that features by now. I guess you have understood that AI is a bigger pictureand machine learning and deep learning or it's apart. So let's move on and focus our discussionon machine learning and deep learning the easiestway to understand the difference between the machine learningand deep learning is to know that deep learning is machinelearning more specifically. It is the next evolutionof machine learning. Let's take fewimportant parameter and compare machine learningwith deep learning. So starting withdata dependencies, the most important differencebetween deep learning and machine learning isits performance as the volume of the data gets increasedfrom the below graph. You can see thatwhen the size of the data is small deep learning algorithmdoesn't perform that well, but why well, this is because deeplearning algorithm needs a large amount of datato understand it perfectly on the other handthe machine learning algorithm can easily workwith smaller data set fine. Next comes the hardwaredependencies deep learning. Are heavily dependenton high-end machines while the machine learning algorithm can workon low and machines as well. This is because the requirement of deep learningalgorithm include gpus which is an integral part of its working the Deep learningalgorithm requires gpus as they do a large amount of matrixmultiplication operations, and these operations can only be efficientlyoptimized using a GPU as it is built for this purpose. Only our third parameter will be feature engineering wellfeature engineering is a process of putting the domain knowledgeto reduce the complexity of the data and make patterns more visibleto learning algorithms. This process is difficultand expensive in terms of time and expertise in caseof machine learning. Most of the features are neededto be identified by an expert and then hand codedas per the domain and the data type. For example, the features can be a pixel value shapestexture position orientation or anything fine the Performanceof most of the machine learning algorithm depends on how accuratelythe features are identified and extracted whereas in caseof deep learning algorithms it try to learn highlevel features from the data. This is a very distinctive partof deep learning which makes it way ahead of traditional machine learningdeep learning reduces the task of developing new featureextractor for every problem like in the case of CN n algorithm it first tryto learn the low-level features of the image such asedges and lines and then it proceedsto the parts of faces of people and then finally tothe high-level representation of the face.
I hope that thingsare getting clearer to you. So let's move on ahead and seethe next parameter. So our next parameter isproblem solving approach when we are solving a problem using traditionalmachine learning algorithm. It is generally recommended that we first breakdown the problem into different sub partssolve them individually and then finally combine themto get the desired result. This is how the machine learningalgorithm handles the L'm on the other handthe Deep learning algorithm solves the problemfrom end to end. Let's take an exampleto understand this suppose. You have a taskof multiple object detection. And your task is to identify. What is the object and where itis present in the image. So, let's see and compare. How will you tackle this issue using the conceptof machine learning and deep learning startingwith machine learning in a typical machinelearning approach. You would first divide the problem into two stepfirst object detection and then object recognization. First of all, you would use a boundingbox detection algorithm like grab cut for exampleto scan through the image and find out allthe possible objects. Now, once the objects are recognized you would useobject recognization algorithm like svm with hogto recognize relevant objects. Now, finally, when you combine the resultyou would be able to identify. What is the object and where it is presentin the image on the other hand in deep learning approach youwould do Process from end to end for example in a euro net which is a typeof deep learning algorithm. You would pass an imageand it would give out the location alongwith the name of the object. Now, let's move on to our fifth comparisonparameter its execution time. Usually a deep learningalgorithm takes a long time to train this is because there's so many parameter ina deep learning algorithm that makes the training longerthan usual the training might even last for two weeksor more than that. If you are trainingcompletely from the scratch, whereas in the case of machinelearning it relatively takes much less time to train rangingfrom a few weeks to few Arts. Now, the execution timeis completely reversed when it comes to the testingof data during testing the Deep learning algorithmtakes much less time to run. Whereas if you compare itwith a KNN algorithm, which is a type of machine learning algorithm the testtime increases as the size of the data increase last but not the least wehave interpretability as a factor for comparisonof machine learning and Running this factis the main reason why deep learning is stillthought ten times before anyone usesit in the industry. Let's take an example suppose. We use deep learning to give automated scoring two essaysthe performance it gives and scoring is quite excellent and is nearto the human performance, but there's an issue with it. It does not reveal whitehas given that score indeed mathematically. It is possible to find out that which node of a deepneural network were activated but we don't know what the neuronsare supposed to model and what these layers of neuronwe're doing collectively. So if able to interpret the result on the otherhand machine learning algorithm, like decision tree gives usa crisp rule for void chose and watered chose. So it is particularly easyto interpret the reasoning behind therefore the algorithmslike decision tree and linear or logistic regression are primarily used inindustry for interpretability. Before we end this session. Let me summarize things for you machine learning usesalgorithm to parse the data learn from the data and make informed decision basedon what it has learned fine. Now this deep learningstructures algorithms in layers to createartificial neural network that can learn and make Intelligent Decisionson their own finally deep learning is a subfieldof machine learning while both fallunder the broad category of artificial intelligencedeep learning is usually what's behindthe most human-like artificial intelligence. Well, this was all for today's discussionin case you have any doubt feel free to add your queryto the comment section. Thank you. I hope you have enjoyedlistening to this video. Please be kind enough to like it and you can comment anyof your doubts and queries and we will reply.
Created by fazeel Ahmed