00:00Hello everyone, welcome to the lecture series on foundations of deep learning concepts and
00:05applications. Today let us discuss on neural networks. Before we get into let us see in the
00:13previous session you have learned about history and evolution of deep learning. We have seen the
00:20importance of computational efficiency in deep learning. You have also seen the overview of
00:26machine learning and deep learning and the key differences between machine learning and
00:31deep learning with example. And today in this lecture we will be studying the most fundamental
00:39building block of deep learning which is a neural networks. Here are the different contents
00:46that we are going to see as part of this session. We will first see the biological inspiration
00:52for neural networks, the mathematical representation of a neuron, the perceptron and the basic components
00:58of a perceptron, the perceptron learning model, various types of perceptron, single layer perceptron.
01:05We will understand this with a numerical example and finally we will end the session with the
01:11limitations of single layer perceptron. Let us understand before we go into an artificial
01:17neuron, we will understand the working mechanism of biological neuron which is the basic inspiration
01:25for building a neural network. So, when you take a human brain, human brain contains billions
01:32of neurons that are interconnected with each other. So, when they are interconnected that means something
01:38is processed in one neuron and that processed information will be passed to another neuron. If one information
01:46has to pass to an other information, we require a channel through which the information can be passed,
01:52that channel we call it as an axon. And then and if this axon is carrying the information,
01:59it carries the information through electrochemical signal. So, let us look into all of this into an
02:06diagrammatic way here. So, we have an input which will enter into the neuron. So, the input will enter
02:15through dendroids. Dendroids is through dendroids is a place through which the input will enter.
02:22And where the input will enter? The input will enter into an nucleus. And what is the main work of
02:29this
02:30nucleus? Inside this there is some processing that is happening. And what is that processing? It
02:37integrates all the inputs which is coming into it. It combines all of this input. And once it combines,
02:45we have a cell body which is the entire portion, we call it as a cell body. And inside this
02:55we have
02:55an nucleus which is responsible for the integration of all of the input which is coming from a dendrite.
03:03And once the information is being processed, the information is being transmitted through an axon
03:14and it carries an electrochemical signal ok. Once we understand this, let us now look into what exactly
03:22is this axon? And if you look into the structure of an axon, axon is an thin fiber like projection.
03:31And the terminal connect, the axon terminal connect to another neuron. We have an axon and the terminal of
03:42the end of this is connected to an another neuron. So, that axon can carry the information and it can
03:48pass
03:49from one neuron to another neuron. But here we have a small problem. What is that? Can you see from
03:55one
03:56neuron to another neuron, there is a small gap. So, when there is a small gap, ideally the information will
04:03not pass. So, that is where the role of synapse will come into. What is an synapse? It is an
04:11connection
04:12point between the axon terminal of one neuron and the dendroids of another neuron. Now, axon is the one
04:22which is carrying the electrical signal. Now, when electrical signal cannot pass because there is a
04:27gap here. So, what happens when it reaches to an synapse? Synapse will convert this electrical signal
04:34signal into an chemical signal into an chemical signal. Now, so that it can pass through this gap.
04:40Now, this chemical signal we call it as an neurotransmitters. So, if the information that
04:48is passed through this axon, if the strength of this signal is very large or very high,
04:55then synapse will release the more neurotransmitters. So, which means there can be more neurotransmitters
05:03that can be released or it can be a less neurotransmitters that can be released.
05:07When more neurotransmitters are released, we say that the signal is stronger. When less neurotransmitters
05:14are released, we say that the signal is weaker. Therefore, the synapse is the one which decides
05:21how much of a message has to get passed to a next neuron. So, I think we have made it
05:29you have
05:30understood or you may be you would have recalled the basic working mechanism of biological neuron.
05:36So, with this let us go into how with this we have built an artificial neuron. So, first let us
05:44go step
05:45by step. First, there is an input which is entering into the nucleus. So, that input is entering through
05:53a dendroids. So, that dendroids part equivalent is the input. Now, from where this input comes?
06:01From our data set, we will have let us take a structured data CSV file, we have a features.
06:07These features are our input. Next, we have inside we have this nucleus. So, I must be telling one point,
06:16this is an important point here. Now, we have seen synapse is one which will decide how much of
06:24information has to be passed to a next neuron. So, which means from the previous neuron, the
06:30information would have entered inside this cell body. So, what exactly happens inside this nucleus is
06:39when a neuron receives an enough input signals through dendroids, it is a cell body. So, it is a cell
06:47body
06:47which is in a biological one we call it as soma. This is the one which decides whether a total
06:55signal
06:56is a strong enough that you can trigger a response.
Comments