00:00neural network. Now, we will keep on hearing this term artificial neural network in all our
00:08subsequent lectures and that is the building block of deep learning. So, what is this artificial
00:15neural network? It is nothing but a mathematical model of the neurons or the network of neurons
00:23that are present in human brain and it mimics how activities happen in human brain. The similar
00:32things are mimicked in artificial neural network. It is inspired by human brain and what is the most
00:40significant part of this deep learning? It can handle large and complex data sets including image
00:48data, speech data, text data. And if we want to see where exactly this deep learning comes
00:58in the overall umbrella of artificial intelligence, we can see here that artificial intelligence
01:04is the main umbrella under which there is machine learning as a subset, under which deep learning
01:09is a subset of machine learning and within deep learning generative AI is another subset of you
01:16know deep learning. Now, we would like to see what is the difference between machine learning
01:21and deep learning? Why we needed to go to that one? Now, for understanding that one, let us take
01:28a problem statement like an image classification problem where you have the image of a puppy and
01:35you want to classify it as a dog or a cat. So, that is the problem statement we are having
01:40in hand.
01:40Now, we will try to see how machine learning approach deals with this problem and how in deep learning
01:47what exactly happens. In machine learning, human first extract features from the image like the
01:55colour histogram, age detection, texture patterns. That means, a human will see what are the significant
02:03features of a dog. Like it will have you know this long kind of ears, the eyes will look like
02:11this,
02:12then it will have a furry texture and then there will be some pattern of colour. So, colour may be
02:18a
02:19little you know a brownish in the face, little lighter you know in the neck portion whatever. So,
02:26there are features which are handcrafted by human. Whereas, in deep learning approach,
02:35we have something called as the convolution neural networks which we will study in our subsequent
02:41sessions. The convolution neural network is trained directly on the raw images with the pixels giving
02:49information about the colour and intensity that is present in that pixel. Now, once in machine
02:57learning approach what happens is, once human extract the features that ok, these are the things which
03:06should be there for an image to be a dog's image. These features are fed into an algorithm like support
03:14vector machines or random forests and then what happens is, from there it does the classification.
03:22Whereas, in deep learning approach, the network automatically learns hierarchical features.
03:28Hierarchical features means, it will first learn to detect the edges like the edges of the dog's image.
03:39Then, it will detect ok, this is an eye, this is a nose, this is the ear, this is the
03:44tail,
03:45this is a furry pattern and finally, it will identify it as a cat or a dog. So, in neural
03:51networks,
03:52particularly in convolutional neural network, what happens is, the first layer, there are multiple
03:56layers. In the first layer, it detects the edges. The middle layers, it detects the eyes, furs,
04:03patterns etc. And finally, it identifies whether it is a cat versus dog, whether it is a cat or a
04:11dog.
04:13Now, the machine learning approach will do very well if the features extracted by human are well done.
04:23So, there is a dependency of human in the overall result that we are getting from the model.
04:31Whereas, in deep learning approach, no need for manual feature extraction,
04:36model learns everything from the data and it is much more automatic. So, here the expected results
04:42are much higher because it is automatically the layers are learning.
04:48Now, finally, we will come to the applications of deep learning.
04:53There are various applications and in today's world, we cannot think lot of things without deep
05:01learning. So, coming to the various applications of deep learning, there is self-driving cars,
05:07language translation, automatic text generation, speech recognition, facial recognition, virtual
05:13assistance, fake news detection, image captioning, all these are just a small subset of where all deep
05:19learning can be used. So, with this one, I come to the end of today's session. So, let us have
05:26a quick
05:27summary of what all we have learnt in today's session. So, we started our session with a brief history
05:33of the evolution of AI and how computers evolution also helped in progressing AI research. Then we looked
05:43into what is machine learning, how it is different from traditional programming approaches and then
05:49we saw the various machine learning techniques, supervised unsupervised reinforcement learnings
05:55and the different algorithms class coming under supervised like classification regression,
06:01unsupervised clustering and you know reinforcement learning which is driven by the feedback from the
06:09environment. Then we saw what exactly happens in supervised learning, what are labeled data set,
06:19how the training happens in supervised learning, in unsupervised how clustering happens, what type of data
06:26is given in unsupervised learning and what exactly is reinforcement learning. Then we looked into the various
06:34applications of this machine learning technique like machine learning regression, for what kind of
06:41applications we use regression like weather forecasting or market prediction. Classification helps in diagnosis,
06:49medical image classifications, in any classification work. And then we saw unsupervised learning,
06:56it helps in lot of recommended systems and customer segmentation based on whatever data they have ok.
07:10And then in reinforcement learning, we have seen that whether they can be used or actually they are used in
07:17gaming AI and robot navigation and other stuff ok, decision making, real time decision making applications.
07:24So, then finally, we came into deep learning and we saw how deep learning is different from machine learning
07:35and what are the different applications of deep learning, why it has such a significant take
07:43in the advancement of technology. So, we saw the difference between deep learning and machine learning
07:49learning and looked into some handful of applications that are that we see in our regular days which are
07:58helped by deep learning, helped by deep learning.
07:59We saw the example.
07:59of machine learning
07:59in
Comments