00:00Hello everyone, welcome to the lecture series on foundations of deep learning,
00:04concepts and applications. This is the first lecture in our series and today we will begin
00:10with a brief look at the history and evolution of artificial intelligence and deep learning
00:15over a period of time. We will also look into how the technological improvement in computers
00:21hardware part helped in the evolution of AI. After that we will go to an overview of machine
00:28learning, we will first discuss about the distinction of machine learning from traditional
00:32programming approach followed by an exploration to the various machine learning techniques
00:37and their applications in solving real world problems. Subsequently, we will move towards
00:43deep learning, examining how it is different from machine learning approach and why it has become
00:49such a transformative technology. By the end of this session, you would be able to appreciate
00:55the significance of deep learning and its wide-ranging applications in advanced technology.
01:01Without further ado, let us go to the overview of machine learning and deep learning.
01:06I will move into the contents for today's lecture. We will start with history and evolution of AI
01:13and deep learning. Then go to introduction to machine learning, difference between traditional
01:17programming approach and machine learning approach. Type of machine learning techniques we will explore
01:23under which comes supervised learning, unsupervised learning and reinforcement learning. Then we will
01:29go to introduction to deep learning, difference between machine learning and deep learning and
01:34finally, applications of deep learning. So, let us start with the history of AI and deep learning.
01:41The AI, the journey of AI started some 80 years back. It started from 1943 when Makulak and Pitt
01:51created the first artificial neuron planting the seed of AI. Then came the Turing test in 1950
01:59where Alan Turing asked the famous question, can machines think? Soon after in 1950s to 1960,
02:08we entered the first golden age of AI when the perceptron and the adaptive linear neurons were introduced
02:15and researchers believed machines could soon learn like human. But then came a setback, the XOR problem
02:23in 1969, which proved that simple network cannot solve complex tasks. So, from that time began the AI's
02:33first dark age. Still, the dream was having died. And in 1980s, with the idea of backpropagation,
02:42it led to the second golden age. Models like neocognitrons and CNNs emerged, showing that computers
02:50could recognize patterns and even read handwritten numbers. In 2000, from 2000 onwards, it saw a steady
02:59progress. And in 2012, AlexNet changed everything. It could recognize images better than ever before,
03:07kicking off the deep learning revolution. And from that 2017, the transformer models were introduced.
03:14And from there, it came GPT-3, ChatGPT, GPT-4, which could understand and create human-like text.
03:21And now in 2025, with DeepSea Car 1, we are actually standing in the third golden age of AI.
03:29Now, we will see how the evolution of computer hardware also helped in the evolution of AI.
03:35In the earlier eras, the processing speed and the CPUs that were available, the memory sizes that were
03:43available with computers were low. And that is the reason only very basic AI algorithm processing was
03:49possible. But in today's technology, due to the advancement of processors, due to the, you know,
03:57a lot of memory available for processing, the AI research is going in a much faster pace.
04:04As hardwares improved, faster processors, large memory made it possible to handle more complex
04:10algorithms and bigger datasets. So, this rise of GPUs and TPUs allowed machines to train deep
04:18learning models much faster. Then we will come to what is machine learning? Machine learning is a subset
04:24of AI that enables systems to learn from data. Machine learning is the first field of study
04:30that gives computers the ability to learn without being explicitly programmed.
04:35Now, what we will see is, what is the difference between machine learning and
04:41traditional programming approach? We will see if there is a task that is given to us,
04:46how traditional programming approach will solve that problem and how this problem will be solved
04:54from machine learning approach. Suppose we take a problem statement like whether a patient is having
05:01ovarian cancer or not. So, this is the problem statement that we have in hand. Now, in traditional
05:07programming approach, the diagnosis will be based on a fixed number of rules. So, if tumor size is
05:16greater than 5 centimeter and CA level is CA 125 level is greater than 35 and irregular shape on ultrasound,
05:23it is likely to be cancer. That means, these rules, there are a set of fixed rules which governs
05:30that whether a patient is having ovarian cancer or not. But what happens is, these symptoms and test
05:39results vary widely between patients. And it is impossible to write down all the possible rules
05:46to detect it accurately. It is a limitation. Now, we will come to see how machine learning approach
05:54handles this kind of problem statement. So, in machine learning, what happens? We train a machine
06:00learning model using medical data from many patients. So, in that medical data, what are the things
06:07that we look for? We take imaging data like ultrasound, MRI, blood test markers like CA 125,
06:14the patient's age, the patient's diagnosis, cancer or not cancer, the previous history, all these things
06:20we will take as you know as a patient data. And a machine learning model will be trained on the
06:29data
06:29that are common among patients who had cancer. So, that when a new patient's data is given,
06:35the model can predict the likelihood of cancer. So, this is the main difference where we are not setting
06:42a set of rules like traditional approach. But in machine learning, we are taking the data of
06:48patients coming from various demography, age with various history of illness. And then,
06:56we are training a model with all this data so that the machine gets the training. And based on that
07:03training, when a new patient's data is given, the model predicts whether there is a likelihood of
07:09having cancer or not. Then, we come to types of machine learning technique. In machine learning technique,
07:17we have three major categories of algorithms, supervised learning, unsupervised learning and
07:24reinforcement learning. In supervised learning, we have classification and regression. In unsupervised
07:31learning, we have clustering. And in reinforcement learning, it is all about learning from the
07:37feedback from the environment. We will see little more in detail in the coming slide.
07:42So, we come to supervised learning. In supervised learning, we train a machine learning model with
07:49something called as labelled data set. So, suppose consider we have this set, you know a rectangle
Comments