Why we need Machine Learning ?

Most beginners In Machine Learning Have this question in mind and even if you are a experienced developer then also this question might comes into your mind ,Let’s try to answer this question using  problems at which machine Learning systems are good at and traditional software development simply fails.In this Article we will first understand about the problems then try to answer those problems using traditional software development then using machine Learning .

The Problems which Machine Learning tries to answer :

  1. Classification i.e dividing the data into classes depending upon the content of data simple example might be classification of images into cat or non cat images.
    A Cat image

    A Non-cat image
  2. Predicting a Value based upon previous results i.e regression i.e predicting house prices based upon all previously available house prices.

    Best fit Line for house price prediction
  3. Generating Similar Type of content based upon input i.e Gan’s

    A Gan’s Generated image
  4. Grouping similar types pf objects i.e clustering

    Similar Data points grouped into one cluster

So, our first problem,Classifying a image as cat or non-cat images ,After applying first principles of problem solving we can see that if we somehow store the pixel values of each pixel of Let’s say image ‘A’ and on every input we compare the pixel values we can say that if all pixel values matches then we can say it is same image ‘A’ or not but there exist a very big problem in this solution that we cannot generalize this solution i.e this procedure is not possible for every image  that ever exist or may exist in future .In that case this procedure cannot be applied in production ,there also exist some methods like histogram of gradients or Fourier transformation of images but they are also applied Maths solutions.Now we with this simple example we can understand why we cannot apply traditional Techniques in solving these problems.

Now Our second Problem,Predicting prices of the houses based upon all previous data of house prices,This problem is fairly solvable but for simple parameters only Like one can come up with a solution that increase in area of house increases the price of houses by a certain variable but let’s be practical This will be a very inaccurate decision as price of the house depends on many more factors like location ,type ,architecture etc and we don’t have any techniques in traditional software development practices  by which we can consider all parameters at once and predict a value without using applied Maths.

Our 3rd Problem i.e Generating similar type of content or i would say just generating some content Like images or Audio ‘s .This has been solved in recent years only Not before that And doing this thing was not possible before ,and forget Software development even traditional machine Learning algorithms was not able to solve this problem,  Solution to this Problem are Generative Adversarial Network’s aka Gan’s ,this is phd work of Ian Goodfellow And his team which includes one of the best scientist in Deep Learning Yoshua Bengio .

Now the last problem Grouping Similar type of Data Points i.e clustering, For basic grouping Like on the basis of Distance is possible with Normal first principle approach and this technique is called K-means Clusternig but doing this without using any sophisticated distance evaluation method is not possible which also includes applied maths .

After Observing all the above result we can easily understand that why do we need Machine Learning ,And by definition Machine Learning is all about applied Maths .If now You are interested in learning Machine Learning check out MachineLearningBootcamp .

 

Send a Message