YouTube2Text

AI Vs ML Vs DL for Beginners in Hindi — Transcript

by CampusX · 2,855 words · 437 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Hey guys, welcome to my YouTube channel
  2. 0:02. This is day two of 100 Days of
  3. 0:04Machine Learning. In this video, we're
  4. 0:07going to talk about a different topic:
  5. 0:11AI vs. ML vs. DL. Okay? So, I've always
  6. 0:13noticed that whenever beginners enter
  7. 0:16this field, this is the first curiosity
  8. 0:18they have. They ask, what's the
  9. 0:20difference? What is AI? What is machine
  10. 0:22learning? And what is DL? And if you
  11. 0:25Google it, if you Google AI vs. ML vs.
  12. 0:27DL, then mostly you'll get this picture
  13. 0:30where you'd see three circles, the
  14. 0:33outermost circle representing AI, and
  15. 0:35inside that, there's a circle
  16. 0:37representing ML, and inside that,
  17. 0:40there's a circle representing DL. Right
  18. 0:43? Now it's clear that the outermost set
  19. 0:46, the biggest thing, is AI. Within AI,
  20. 0:49a subset is machine learning, and
  21. 0:51within machine learning, a subset is
  22. 0:53deep learning, which is actually
  23. 0:55correct. But what I want to give you is
  24. 0:57a very good intuition as to why this is
  25. 1:00so. Okay? So let's discuss that. Let's
  26. 1:03first discuss AI. AI stands for
  27. 1:05artificial intelligence, or simply,
  28. 1:08infusing intelligence into machines.
  29. 1:11Right? And this idea has been around
  30. 1:13for a long time. Probably for 100 years
  31. 1:15. We humans have been wondering, what
  32. 1:18if we could make machines as
  33. 1:19intelligent as we are? Right? Now, if
  34. 1:22you think carefully, what is
  35. 1:23intelligence? Right? We humans are
  36. 1:26better than other biological species,
  37. 1:28we are advanced, we are evolved.
  38. 1:31Because of our intelligence. Right? But
  39. 1:33what exactly is intelligence? So, if
  40. 1:35you study and research intelligence a
  41. 1:37little, you'll understand it's a
  42. 1:39complex thing. And it's made up of
  43. 1:41multiple things, right? So, there's
  44. 1:43pattern recognition, which we use on a
  45. 1:45day-to-day basis. Be it coding or
  46. 1:47solving puzzles. These kinds of things
  47. 1:50are what you might call IQ. It also
  48. 1:52includes creativity and imagination. It
  49. 1:55also includes emotional intelligence.
  50. 1:57Right? So, our intelligence is a very
  51. 1:59complex thing, encompassing many other
  52. 2:02things. Right? But if you talk about AI
  53. 2:04, at this point, the AI we human beings
  54. 2:07are pursuing is a subset of actual
  55. 2:09intelligence. We're trying to create a
  56. 2:12general intelligence that behaves
  57. 2:14exactly like us. But it's actually very
  58. 2:16difficult. So, what we people are
  59. 2:18trying to do is we're trying to go a
  60. 2:20specific path, and that is pattern
  61. 2:22recognition. So, most of the AI stuff
  62. 2:24you'll see around here is mostly
  63. 2:26related to pattern recognition. We
  64. 2:28can't go for creativity and imagination
  65. 2:30at this point, and emotional
  66. 2:32intelligence at this point because
  67. 2:34that's hard to quantify. We don't know
  68. 2:36the true results. We don't even know
  69. 2:38what the equation of love is yet, or
  70. 2:40how to measure creativity. Right? So,
  71. 2:43for us, what's measurable, what's
  72. 2:44quantifiable, is pattern recognition.
  73. 2:47And that's why, at this point, whatever
  74. 2:49you see in AI is mostly related to
  75. 2:51pattern recognition. Okay? Now, let's
  76. 2:54talk about AI. Let's take a look at all
  77. 2:57the work that's been done so far. So,
  78. 2:58this idea started floating around
  79. 3:00around the 50s. At that time, Alan
  80. 3:02Turing and all those people started
  81. 3:03thinking about whether intelligence
  82. 3:04could be instilled in machines. And at
  83. 3:07that point, intelligent systems began
  84. 3:11to emerge. In fact, at that time, a new
  85. 3:14wave of intelligence emerged, a new
  86. 3:16wave of AI called symbolic AI. What
  87. 3:18happens there? What do you do in
  88. 3:20symbolic AI? You build a massive
  89. 3:22knowledge system using if-else
  90. 3:24conditions. Basically, you write down a
  91. 3:26logic for each condition. Right? And
  92. 3:29from that symbolic AI, expert systems
  93. 3:31began to emerge. For example, if you
  94. 3:34ever played chess against a computer as
  95. 3:36a child, that was actually an expert
  96. 3:38system where an expert was seated, all
  97. 3:41the knowledge was extracted from him,
  98. 3:43converted into code, and then we play
  99. 3:45chess against that code. Right? Now,
  100. 3:48what happens in expert systems? Expert
  101. 3:50systems have a knowledge base that we
  102. 3:52take from an expert, and then there's
  103. 3:54an influence engine, a program that
  104. 3:56makes decisions based on that knowledge
  105. 3:58base. Right? So in the 80s, everyone
  106. 4:00started believing that expert systems
  107. 4:02were the future of AI, and AI and
  108. 4:04expert systems were synonymous at that
  109. 4:06point in time. But later, people
  110. 4:08realized that expert systems have some
  111. 4:10flaws. What are those flaws? Expert
  112. 4:12systems work on very closed problems.
  113. 4:15Meaning, on very specific problems like
  114. 4:18chest detection or, I don't know, lung
  115. 4:20detection, for medical purposes. But if
  116. 4:23you give it a problem that is a little
  117. 4:25fuzzy, whose logic is a little fuzzy.
  118. 4:27It is called fuzzy logic, for example,
  119. 4:30I don't know, there is a dog in an
  120. 4:32image. There is a dog, it has to be
  121. 4:34detected. Now, this is fuzzy logic
  122. 4:36because there are so many breeds of
  123. 4:38dogs. And all of them have different
  124. 4:40features. So if you sit down to write
  125. 4:42if else, you can't write all the
  126. 4:45conditions so as to build an expert
  127. 4:47system which can actually classify dogs
  128. 4:50in an image. Right? So, for such
  129. 4:52problems where fuzzy logic was used,
  130. 4:54image classification was done, voice
  131. 4:56recognition was done, and apart from
  132. 4:58this, there were many other areas where
  133. 5:00this symbolic AI logic was not
  134. 5:02applicable. And that is why expert
  135. 5:04systems were left behind, and then came
  136. 5:05machine learning, which solved this
  137. 5:07problem. So, what we are discussing
  138. 5:09next is machine learning. So now let's
  139. 5:11talk about ML. Where does ML fit in? So
  140. 5:14if you have seen the last video, there
  141. 5:16I told you the definition of machine
  142. 5:17learning. Machine learning is a branch
  143. 5:20of computer science where you use
  144. 5:23statistical techniques to find patterns
  145. 5:25in your data, or more effectively, we
  146. 5:28can say that the technique of learning
  147. 5:31patterns in data is called machine
  148. 5:33learning. Right? The most important
  149. 5:36difference between machine learning and
  150. 5:38the symbolic AI that was previously
  151. 5:40used is that in machine learning, you
  152. 5:42never use explicit programming.
  153. 5:43Explicit programming means rule-based
  154. 5:45programming. You write logic for
  155. 5:47everything. Right? You don't do that in
  156. 5:49machine learning. Instead, what you do
  157. 5:51is you have data. Data has both inputs
  158. 5:54and outputs. You feed that data to your
  159. 5:56system, and your system's job is to
  160. 5:58derive rules or patterns. To derive
  161. 6:00mathematical patterns. So, when you get
  162. 6:03new input, now that you know the rules,
  163. 6:05you can produce a new output or
  164. 6:07prediction. Right? Now, why is this
  165. 6:09approach better than symbolic AI? This
  166. 6:12is very easy to understand. Where was
  167. 6:14symbolic AI failing? In fuzzy logic, in
  168. 6:17fuzzy logic-based problems. For example
  169. 6:19, in an image classification problem
  170. 6:21where you have to classify dogs. Now,
  171. 6:23you can't write rules to identify every
  172. 6:27dog breed. Right? But in machine
  173. 6:30learning, you don't have to write rules
  174. 6:31. Your system will generate the rules.
  175. 6:34All you have to do is provide the
  176. 6:35correct data. So now what I'll do is
  177. 6:37I'll provide my system with a lot of
  178. 6:40photographs of a lot of different dog
  179. 6:42breeds. Right? I'll show a lot of
  180. 6:44photographs. I'll tell it, "This one is
  181. 6:45a dog. This one is not a dog. This one
  182. 6:47is a dog. This one is not a dog." Now,
  183. 6:48what the system will do is it will
  184. 6:50automatically try to find the
  185. 6:52underlying representation of the data.
  186. 6:54This is called learning. It's trying to
  187. 6:56learn patterns, to recognize them, and
  188. 6:58that's what's called learning. And if
  189. 7:00you show it a lot of photos, eventually
  190. 7:03it will be ready to classify whether a
  191. 7:06given image has a dog in it or not.
  192. 7:09Right? So that's the main idea. It's a
  193. 7:11simple one. The idea is, why write
  194. 7:13rules? Show them data, and your system
  195. 7:16will automatically understand what's
  196. 7:17considered a dog and what's not. In a
  197. 7:21way, we work like this. As children,
  198. 7:25we're also taught, look, this is a dog,
  199. 7:25this isn't. Right? And that's how we
  200. 7:27learn. So, in a way, that learning is
  201. 7:29inspired, in part, by our own evolution
  202. 7:32of intelligence. Right? So this was
  203. 7:34machine learning. And guess what? The
  204. 7:36simple idea that rather than writing
  205. 7:38rules for everything, you can simply
  206. 7:40provide data, and the rules will be
  207. 7:42generated automatically has changed the
  208. 7:44industry. In the last 30 years, machine
  209. 7:46learning may have come in the 90s, it's
  210. 7:48only 2020. It has been just 30 years,
  211. 7:50and this thing has been revolutionary.
  212. 7:52Especially in the last 15, 10 years,
  213. 7:54when data is so abundantly available,
  214. 7:57and good hardware is available, machine
  215. 7:59learning has taken off and is flying
  216. 8:01high. Right? Many people used to say,
  217. 8:03in fact, that machine learning is
  218. 8:05nothing but statistics. But to be
  219. 8:07honest, it's not purely statistics. Yes
  220. 8:09, we are using statistical tools. We
  221. 8:11are using mathematical tools. But there
  222. 8:13isn't much mathematics involved. There
  223. 8:15is mathematics only to a very limited
  224. 8:16extent. It's highly practical
  225. 8:18engineering-based. Okay? So, it's
  226. 8:21inspired by statistics, but not exactly
  227. 8:23derived from it. Okay? So, I guess you
  228. 8:25understand machine learning. Now comes
  229. 8:27the question: what is deep learning? If
  230. 8:30machine learning could do everything so
  231. 8:32well, why is deep learning needed?
  232. 8:34Right? Let's discuss that. So, let's
  233. 8:37discuss deep learning in two parts.
  234. 8:39First, let's discuss what deep learning
  235. 8:41is, and then let's discuss why deep
  236. 8:43learning is needed. Okay? So, deep
  237. 8:46learning is very recent. Like, already,
  238. 8:48work on it has been going on for a long
  239. 8:50time. But it wasn't until 2012 that we
  240. 8:52reached the peak of deep learning,
  241. 8:54which is why we're hearing so much
  242. 8:56about it now. Okay? So, deep learning
  243. 8:59is basically a class of algorithms.
  244. 9:02First, it's machine learning. It's not
  245. 9:04different. Meaning, never have the
  246. 9:06notion in your mind that deep learning
  247. 9:07is something else. You use the same
  248. 9:09procedures, the same process flow, in
  249. 9:11deep learning as you do in machine
  250. 9:12learning. You give the same data, train
  251. 9:15the algorithm, and reduce the error
  252. 9:17during training. Eventually, your model
  253. 9:20is trained. Then, you give it new data
  254. 9:22and make predictions on it. The exact
  255. 9:23same process occurs in machine learning
  256. 9:25. The exact same process occurs in deep
  257. 9:26learning. So, deep learning is
  258. 9:27essentially machine learning. It's just
  259. 9:29that the algorithms you use here are
  260. 9:31slightly different. And in a way, you
  261. 9:34can say that it's inspired by biology.
  262. 9:36Neurons are inspired by our biology. I
  263. 9:38wouldn't say that it works completely
  264. 9:40like our brains. Deep learning is a
  265. 9:42different concept. Deep learning is
  266. 9:44inspired by our biology. But we can't
  267. 9:47say that deep learning algorithms work
  268. 9:49exactly like our brains. To be honest,
  269. 9:52we don't yet understand how our brains
  270. 9:53work. Deep learning is a mathematical
  271. 9:56model. It's math-based. It's just that
  272. 9:58its core unit, the smallest unit, is
  273. 10:00called a neuron perceptron. It's based
  274. 10:02on biological neurons. So that's why
  275. 10:04there's this biological connection.
  276. 10:06Many people are confused that deep
  277. 10:08learning works the way the brain does.
  278. 10:10It doesn't. Okay? So that's about deep
  279. 10:12learning. So why did deep learning come
  280. 10:15into use? Or why? Deep learning is so
  281. 10:17popular today. It's such a buzzword in
  282. 10:19the industry. The reason is that there
  283. 10:21are certain things that machine
  284. 10:23learning can't do, or you can say,
  285. 10:25can't do as well, and that's where deep
  286. 10:27learning shines. Okay? Let me tell you
  287. 10:29where. So, a big problem with machine
  288. 10:31learning is that in machine learning,
  289. 10:33you have to provide features. You have
  290. 10:35to tell the input you provide what the
  291. 10:37input will contain. For example, if you
  292. 10:41want to classify between dogs and cats
  293. 10:43in an image, saying, "This one contains
  294. 10:45a dog, this one contains a cat," you'll
  295. 10:48have to specify the features that
  296. 10:50belong to dogs. And you'll have to
  297. 10:51specify the features that belong to
  298. 10:53cats. The ears are pointed. It has to
  299. 10:55be a dog. Right? It has a mustache. It
  300. 10:59has to be a cat. Right? The tail is
  301. 11:00curled up a lot. It has to be a dog.
  302. 11:03Right? This is how you'll have to
  303. 11:04define features. If you're in the
  304. 11:06machine learning space, right? But if
  305. 11:08you're in the deep learning space, what
  306. 11:10happens in deep learning is that your
  307. 11:12system automatically creates or detects
  308. 11:14these features. You don't have to
  309. 11:17create them, right? And if you're
  310. 11:19working with fuzzy logic-based problems
  311. 11:21where you actually don't know what
  312. 11:23features might exist, deep learning can
  313. 11:25be very useful there. I'll give you one
  314. 11:28more example. Like, we were trying to
  315. 11:30build a system in our startup where we
  316. 11:32could look at resumes and decide
  317. 11:34whether someone would be placed or not.
  318. 11:36We were trying to build a system like
  319. 11:38this. Now, if you work in machine
  320. 11:40learning, you'll have to create
  321. 11:42features that you'll feed to the
  322. 11:44machine learning model. You'll have to
  323. 11:46create those features, and you'll only
  324. 11:48be able to do that if you have
  325. 11:50knowledge about the data. Right?
  326. 11:51Someone who knows a lot. Probably,
  327. 11:53someone working in a TMP cell will know
  328. 11:55that if they have more than two
  329. 11:57certifications, there's a chance of
  330. 11:59getting placed. If they have a
  331. 12:01percentage between 90 and 95%, there's
  332. 12:03a chance of getting placed. If there
  333. 12:05are backlogs, there's no chance of
  334. 12:07getting placed. All of these are
  335. 12:09individual features, right? But if
  336. 12:10you're working in the deep learning
  337. 12:12space, what you have to do is simply
  338. 12:14provide the raw text, and it will
  339. 12:15automatically create features. So
  340. 12:17that's one of the big differences
  341. 12:19between machine learning and deep
  342. 12:21learning. In deep learning, you don't
  343. 12:23have to be concerned about the features
  344. 12:25. What deep learning will do is
  345. 12:27automatically extract those features.
  346. 12:30In machine learning, you have to create
  347. 12:31those features. And that's a difficult
  348. 12:34task: understanding which features will
  349. 12:35be more accurate. Right? Second, in
  350. 12:38deep learning, as you add new layers of
  351. 12:44neurons, the more your classification
  352. 12:47or regression task becomes better. Or,
  353. 12:50more simply, as you add layers of
  354. 12:54neurons, the system's efficiency and
  355. 12:57prediction power improves. Because with
  356. 13:01each layer, you can uncover deeper and
  357. 13:04deeper what's hidden in your data. If
  358. 13:06you look at this image, you'll
  359. 13:08understand that gradually, if we're
  360. 13:10trying to classify something, say, a
  361. 13:12digit, the first layer will tell you
  362. 13:14the edges. The second layer will
  363. 13:16combine the digits into bits and pieces
  364. 13:18, and the next layer will combine those
  365. 13:19digits to tell you which digit it is.
  366. 13:21Right? So that's how deep learning
  367. 13:23works. Right? And this is one of the
  368. 13:25main reasons why deep learning was used
  369. 13:28where machine learning wasn't
  370. 13:29performing as well. And deep learning
  371. 13:32gave good results. The second reason is
  372. 13:34that deep learning is something else.
  373. 13:36It's because of the algorithms that the
  374. 13:40more data you feed it, the more its
  375. 13:42performance improves. Right? If you
  376. 13:45look at this graph, you'll understand
  377. 13:47that machine learning models improve
  378. 13:50their accuracy up to a point with
  379. 13:51increasing data, but after that, they
  380. 13:54stabilize. This isn't the case with
  381. 13:55machine learning and deep learning. If
  382. 13:57you keep feeding it more data, your
  383. 13:59performance will continue to improve.
  384. 14:02Right? So, for these two or three
  385. 14:04reasons, deep learning has found a
  386. 14:07niche where it's used a lot, like image
  387. 14:10classification tasks, basically
  388. 14:12image-related tasks, and text-related
  389. 14:15tasks. And these are some of the
  390. 14:17applications where deep learning is
  391. 14:20better than machine learning. But then
  392. 14:23you might ask, why use machine learning
  393. 14:24if deep learning is so much better? The
  394. 14:26answer is that where you need a needle,
  395. 14:29you wouldn't use a sword, right? So,
  396. 14:31deep learning is a sword. It's a
  397. 14:32powerful thing, it's a powerful
  398. 14:34technology, but you wouldn't use it
  399. 14:36everywhere. Machine learning is really
  400. 14:38superior in some areas. You use deep
  401. 14:40learning on big data. You get good
  402. 14:43results. But with small data, you don't
  403. 14:45have a lot of data available. So you
  404. 14:47won't use deep learning because machine
  405. 14:49learning will give you better results
  406. 14:50on less data. And guess what? 90%of the
  407. 14:53world's data is small data. Big data is
  408. 14:56all you use, like Facebook and WhatsApp
  409. 14:58. So, you might think everyone has a
  410. 15:00lot of data. But that's not true. If
  411. 15:01you ever work in industry, you'll find
  412. 15:03that banking companies, insurance
  413. 15:05companies, and sports analytics
  414. 15:07companies don't have a lot of data. So,
  415. 15:09machine learning is still very useful
  416. 15:11in such scenarios. Okay? So, that's the
  417. 15:13complete point. Let me summarize it
  418. 15:15once. The biggest term is AI. We're all
  419. 15:19after AI. Now, the most efficient
  420. 15:21system for creating AI is machine
  421. 15:23learning. Right? And even in machine
  422. 15:25learning, a method has emerged to solve
  423. 15:27some very complex problems, which we
  424. 15:30call deep learning. And in today's date
  425. 15:32both Deep Learning and Machine Learning
  426. 15:34are very famous. Now you never know how
  427. 15:36to build a general AI, after 10 years
  428. 15:37something completely out of the box new
  429. 15:39may come. But if at this point you are
  430. 15:42planning to learn to build AIs, Machine
  431. 15:44Learning and Deep Learning is the right
  432. 15:46path. Okay? So I hope you understood
  433. 15:48this entire video guys. If so please
  434. 15:50consider subscribing. I will try to put
  435. 15:53a new video every day. This is our goal
  436. 15:56100 days of Machine Learning. So or
  437. 15:58that's it. Thanks for watching.

About this transcript

This page contains the full transcript of AI Vs ML Vs DL for Beginners in Hindi by CampusX, generated from the public captions YouTube serves with the video. The transcript has 2,855 words across 437 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.