What is Machine Learning? | 100 Days of Machine Learning

Summary

Highlights

• CampusX is launching '100 Days of Machine Learning' to fill a knowledge gap in end-to-end ML learning. 🎉
• The series is structured for beginners and intermediates, focusing on ML lifecycle over algorithms. 👩‍💻
• Learners will explore topics like data handling, deployment, and bias-variance trade-offs. 🧐
• Real-life applications of ML, like spam detection and image classification, are discussed. 📧🐶
• A historical perspective on ML's rise akin to actor Nawazuddin Siddiqui's journey is provided. 🕰️

Key Takeaways

• CampusX's '100 Days of Machine Learning' is a structured series tailored for beginners and intermediates. 📚
• The series focuses on understanding the machine learning lifecycle rather than just algorithms. 🔄
• Key learning topics include end-to-end ML project development and dealing with practical challenges. 🚀
• Machine learning is explained as deriving patterns from data without explicit programming. 🎓
• ML competes with traditional programming in scenarios like email spam detection and image classification. 📈

Overview

What is Machine Learning? | 100 Days of Machine Learning Transcription

• 00:00 - 00:30 Hey Guys, Welcome to my YouTube Channel In this video, I am going to make an announcement. that, I'm going to create a new Playlist for my YouTube Channel. and I'm going to call that Playlist "100 days of Machine Learning" So... You might be thinking Sounds interesting But what is it exactly? Here's the thing In the past few days, Many of you messaged me.. over WhatsApp or through Youtube comments regarding They weren't able to find.. a end-to-end machine learning playlist on my channel.
• 00:30 - 01:00 and that is true actually. If you go to my channel, you would find videos related to Machine Learning algorithms I've created videos on various machine learning algorithms. But a complete end-to-end playlist isn't available in my channel. So,I thought why not create one comprehensive Playlist on Machine Learning? So, that's the plan. In the coming 100 days, I have planned to shoot and upload one video every day.
• 01:00 - 01:30 I've developed a structured curriculum To the best of my experience and knowledge, I think this Playlist will be more than sufficient to teach you.. Intermediate-level machine learning. For advanced-level machine learning, it's completely in your hands. but if you're a beginner or slightly more experienced, If you follow the course throughout, I hope you'll advance to a proficient level in Machine Learning. Now comes the question about What are the topics we'll be covering?
• 01:30 - 02:00 So, I will tell you something about myself When I began learning machine learning, my primary focus was on mastering ML algorithms. Gradually, I realized that after learning the algorithms and working on few projects, there are two essential aspects to focus on in Machine Learning One of them is learning algorithms knowing about algorithms is mandatory. But along with that, Simultaneously, it's crucial to understand how to develop an end-to-end machine learning project,
• 02:00 - 02:30 including the complete flow. We call this as "Machine Learning Life Cycle" also known as the "Product Life Cycle". and beginners usually doesn't focus much on this aspect. Beginners often believe that knowing Machine Learning algorithms alone is sufficient. but not really. So, I planned in the following 100 days.. I will start with the basics of Machine Learning and I'll cover the entire flow of Machine Learning. If you want to do a Machine Learning project, I'll cover all the potential challenges you might encounter while working on an Machine Learning project.
• 02:30 - 03:00 Since we're covering 100 topics in 100 days, no topic will be left untouched. Keep in mind, we won't be delving into algorithms in this Playlist. We are not going to cover any of the algorithms. For algorithms, we already have a separate playlist in our channel. If you want to learn about any algorithms, then you can go to the playlist of that particular algorithm. But in this "100 days of Machine Learning",
• 03:00 - 03:30 Here, we learn the techniques we learn the flow we learn about how we deploy. We'll learn how to perform imputation, how to perform pre-processing, how to perform analysis, model selection, feature selection, and such important concepts like What is Bias-Variance Trade Off? and the weighted and important topics that differentiates ordinary Machine Learning engineers from extraordinary Machine Learning engineers. We'll cover such topics here.
• 03:30 - 04:00 I am still working on the curriculum it is not completed yet. I will be uploading that curriculum as well in the coming few days. In fact, if want any certain topic to be covered, you can share that with me and I'll be covering that topic as well. But I'm thinking in this 100 days, I have to create a resource that is meaningful for you and your Machine Learning journey. and the other question which may pop out in your mind For whom is this playlist intended?
• 04:00 - 04:30 If you are a beginner, then it is definitely for you. and even if you know intermediate machine learning, it could be a great resource for you as you can come here and use this as a valuable resource to watch and learn anything that you could've missed, or gain a deeper understanding of familiar topics with clarity. This will be helpful for you in such cases. In short, this will benefit everyone following the channel, including students and professionals.
• 04:30 - 05:00 I'll try to be very honest while making these videos I'll try to put my maximum effort So, yeah.. This is the announcement I was planning to make. Now, I am not going waste any time I will start it today itself. Today we are going to cover the first topic which is "What is Machine Learning?" I know that many of you already knew it. However, I want to begin everything from scratch. So let's cover this topic: What is Machine Learning?
• 05:00 - 05:30 So, let's dive into what machine learning is We'll start with the formal definition According to definition, Machine learning is a field of computer science that uses statistical techniques to give computer systems the ability to "learn" with data, without being explicitly programmed. To put it in simpler terms, Machine Learning is all about learning from data. There is it a term in this definition called as "Explicit Programming" Let's explore the concept of explicit programming.
• 05:30 - 06:00 Explicit programming involves writing code for each specific scenario. To handle that scenario, you write a code. But in Machine Learning, you don't do that. what you do is that you've got some data and you've got an algorithm You instruct the algorithm to explore the data and identify patterns between input and output. Once you've identified the patterns, we provide new input to the algorithm to derive the output. If you check out this flow diagram, you can observe that in conventional programming approach
• 06:00 - 06:30 we write a program for which logic is written by us. If you give input to that program, you'll get your output. But in Machine Learning, things are different. what you do is you provide some data. In that data, you give an input as well as an output But you haven't written any program or logic. That logic is generated by none other than the Machine Learning algorithm. The good part is that you don't have to write code for each condition/case.
• 06:30 - 07:00 It is automatically handled by Machine Learning algorithm. For example, you've written code for adding two numbers Whenever you give two numbers to that program, it returns you the value of sum But in Machine Learning, what you will do is you give data You'll give an excel file in which the rows contain the numbers and their respective sum.
• 07:00 - 07:30 Whenever Machine learning models train on that data, The model realises the pattern as addition. After training, irrespective of giving two or four or ten numbers as an input, your machine learning model knows that it has to perform addition. It adds all of them and gives to you. Whereas in the code written for sum of two numbers, If you give more than two numbers as input, then the program doesn't function since it is explicitly coded to perform sum of two numbers.
• 07:30 - 08:00 That's the key difference I hope from the example you can understand the reason behind the powerful nature of machine learning in the industry. Now that we know about Machine Learning, Let's also discuss when and where Machine Learning is used Let us know in which type of scenarios, Machine Learning can be used and is useful than traditional software development I'll provide you three scenarios
• 08:00 - 08:30 and there are other scenarios as well. These three scenarios feel significant to me. First scenario is that you can't perform few things using programming You can't write cases for everything In such situations, we use Machine Learning I will give you one real life scenario Consider that you are trying to build an e-mail spam classifier to detect whether the given email is spam or not. If you were given to write a program for that as a software developer,
• 08:30 - 09:00 Then what would you possibly do? you will pick a bunch of e-mails and you'll have the information about that e-mail, whether is spam or not then you will try to create patterns for them. like any word such as 'discount' or 'sale' or 'awesome' is repeated more often or it is filled with bunch of pictures, then you create a long if-else ladder by using if else for each and every condition That would be the possible program for you Spam classifier.
• 09:00 - 09:30 but let us say that you've written something like if 'huge' is repeated more than three times, huge as in discount is used, you'll label that e-mail to be spam. and somehow advertising companies got to know that if code is written to classify the mail as spam if the world is repeated more than three times
• 09:30 - 10:00 then those companies might use 'big' or 'massive' instead of 'huge' then the program couldn't pick that mail as spam. then you've to modify the logic of the code to handle a new scenario but again, advertising companies can experiment with different words if they get to know about it In short, you've to keep changing the logic more frequently
• 10:00 - 10:30 to make sure that the code runs effectively. In machine learning, it doesn't happen that way since it learns from data, if data changes, then it will automatically get reflected in logic That's the beauty of 'Machine Learning' You just have to write just one algorithm, and everything will be handled by that algorithm itself and second scenario where Machine Learning is useful over traditional programming
• 10:30 - 11:00 A scenario where you cannot even imagine the number of cases like 'Image Classification' Let us say you want to classify dogs, that if dog is present in picture or not There will be hundreds of breeds which varies in looks, few large and few short, vary in colours and other characteristics. So, if you were to create a program that detects the presence of dog in a picture
• 11:00 - 11:30 Can you imagine the number of cases you've to write inorder to cover characteristics of every breed? You cannot do so. You cannot code it We have use the technique that we, humans, use to identify dogs We were thought from our childhood to identify that particular animal is dog That one is not a dog, that is a cat Our mind mentally keeps tagging the name with the animal
• 11:30 - 12:00 It keeps learning from the data It is also one of the scenarios where you cannot use conventional software development approach You'll have to use the Machine Learning approach One more important use case is Data Mining. What exactly is Data Mining? First, let's discuss about what data analysis is. Data analysis is a process where you extract patterns or search for hidden information by plotting graphs
• 12:00 - 12:30 That is 'Data Analysis' but sometimes the information is more hidden which you won't able to get through graphs I'll give you one scenario Just by looking at the e-mail content, if we're unable to detect key words due to which we can the e-mail as spam, then you perform Data Mining. In Data Mining, you apply Machine Learning algorithm on the data
• 12:30 - 13:00 you create a prediction model just like e-mail spam classifier you can check the patterns extracted by the Machine Learning model like if 'huge' is occurring more frequently, then there it is labeled as a spam if it is not that frequent, it is treated as not spam After applying Machine Learning, If you are able to extract important data from the information,
• 13:00 - 13:30 That is known as "Data Mining" Most of the times in order to extract the hidden patterns, in order to perform such data analysis, we use Machine Learning and this is called as Data Mining Machine learning is a very important tool to perform data mining You should have understood the importance of machine learning using these 3 to 4 scenarios And why is it taking over the world Next you should be knowing a little history about Machine Learning
• 13:30 - 14:00 You should be learning about it's history if you are starting on any new technology I feel that history of machine learning is more or less like Nawazuddin Siddiqui I hope you all know about him, he is a great actor Just like Nawazuddin Siddiqui, machine learning is already existing from a long time in the industry
• 14:00 - 14:30 He was playing a very small role in Munna Bhai MBBS Similarly, machine learning is there from 40 to 50 years But it couldn't get into limelight like other important technologies Until the recent 2010s Only from then,it raised to the level that machine learning is today
• 14:30 - 15:00 If you talk about Nawazuddin Siddiqui, what could be the success reason behind him? It might be the OTTs or the audience preferring the content-based films So there is a paradigm shift which led him to be one of the biggest actors of the country. Similar events occurred in the case of machine learning also All the theory and Maths existed from very long time
• 15:00 - 15:30 but machine learning is not that famous because of the reason that machine learning requires significant amount of data Unfortunately,back then gathering and labelling the data is a quite tedious task Also, the inefficiency of hardware to run algorithms on such data back then After 2010, with evolution of internet and smartphone,
• 15:30 - 16:00 These two problems are sorted out. We are generating data at a heavy pace, You can consider the example of your own life from morning to evening before going to bed So much of data is generated by yourself alone Then imagine the data generated by 4 billion Internet users around the globe In fact, the amount of digital data created from the starting of mankind till 2015 is generated in 2016 alone
• 16:00 - 16:30 That is the speed at which we are generating data this data is aiding the growth of machine learning Second thing is hardware In the modern day world, we ourselves are carrying up to 12GB of RAM in our pocket in the form of mobile We are carrying GPUs in our pocket which were not available to research scientists. Even 128MB RAM was a big deal then but we are equipped with good hardware, data and algorithms now
• 16:30 - 17:00 That is the reason why machine learning is enjoying it's fruits This is not going to stop any time soon The growth curve will keep on growing exponentially and that is the reason for jobs in this sector If we talk about jobs, I will also have to discuss about few things Do you think the jobs that are available now, the salaries which we get now in this industry
• 17:00 - 17:30 Will they continue to be same in the future? The answer is NO This is just pure economics When Java entered the market, only a handful of people were familiar with the language. But companies needed Java because their competitors were implementing it Therefore, as a company, I also need Java in my software For that, I have to hire some professionals But when I reach out to market for hiring process, I realised that there is a lack of talent in this space
• 17:30 - 18:00 So all the companies will be fighting for those few available professionals So that would obviously lead to more salary of that professional SIMPLE ECONOMICS The similar trend is now going on with machine learning Even in the colleges, machine learning is not being taught and many of the engineers doesn't know machine learning as of now, which is changing gradually Every time when a company goes to a college, they only find less number of students who know Machine Learning
• 18:00 - 18:30 So they have to fight for those students resulting in a higher salary Over time, as salaries increased, more people became interested in learning the technology to secure those jobs. When everyone in the market learns machine learning in the coming few years
• 18:30 - 19:00 Majority of the population would be knowing machine learning just like Java today Once most of the people knows machine learning, the salaries would be automatically normalised as companies will have a lot more options and that's the reason why they won't be paying high salaries that they're paying today The good part is that we are at an initial growing phase
• 19:00 - 19:30 The graph of any technology will be like this, initially increasing and then decreasing The positive aspect is that for those learning machine learning, we are on an upward trajectory. and there is still time which I feel If we learn correctly now, we can expect to achieve the same level of success that machine learning is currently experiencing. I understand it's been a lengthy video, but this was the introduction I wanted to provide on machine learning.
• 19:30 - 20:00 In the next video, we'll delve into an essential topic that often raises doubts among beginners. What is the difference between AI, ML and DL? We will be covering this on the next day, which is Monday. That's a wrap for this video. I hope you like it. If you plan to follow this series, '100 Days of ML,' please consider subscribing to the channel. Thank you for watching!