Pre Trained Models and Deep Learning with CNN: Know the enhancement of AI 2023

Pre Trained models - Deep Learning and CNN

Welcome guys today we will see, what is Pre trained models and learn about how deep learning with CNN convolutional neural network plays a role in the enhancement of Artificial Intelligence.

Deep Learning – Simply means Data Hungry, it requires more data to perform well.

CNN – Convolutional Neural Network, mostly image recognition process is involved, so if we talk about CNN it requires more labeled image data to perform well.

Why we should use Pre Trained model instead of our own model?

Reason No: 1

As, pre trained models are someone else’s model we can directly import the model into our project, if suppose we start to make a pre trained model from scratch and I’m talking about CNN model, then we will require lots and lots of images, collecting images is not a big task right? We can directly download images from Google. But the twist is that, we need all these images labeled individually to process in CNN model.

Suppose I downloaded 10000 images of trees and plants from Google. We need to label each image as trees and plants respectively. And one person can’t do the task of labeling by himself he will require some help to do this big task, so he must hire some people and give them salary, and all this process may cause financial loss, this is the first reason that we should use pre trained model instead of our own.

Reason No: 2

Here we are talking about our own model and to make this model a pre trained model we must train our model first, right? So to train our model it may take hours, days or maybe weeks according to the provided data. So using a pre trained model which is previously trained by someone can save a lots of time, this is the second reason.

Let’s take a look how pre trained model came into existence

Know how Deep learning with CNN plays a crucial role

So to understand the concept of pre trained model you must know what is, “ImageNET Dataset”

ImageNET Dataset is simply known as visual database of images. Let us dive deep, to know (Why & How) about ImageNET Dataset.

ImageNET dataset

Starting with “Why”, Why ImageNET Dataset was made?

In 2006 there was a computer researcher named Fei Fei Li and in 2006 deep learning was becoming popular and at that time mostly computer researchers were focusing on models and algorithms and Fei Fei Li was one of them. He thought if deep learning is all about data, so in future it is absolutely necessary that we must have some big and cool datasets on which these models or algorithms will work.

So from 2006 he started building this database, “ImageNet Dataset”. He met with a professor who made a dataset for words called “WordNET Dataset” and he built the database of images using 140 million images. Not only this, it was divided into 20,000 categories. For example, man, women, vehicle, cat, dog, etc. Those each images were well labeled, like take dog as an example, it will say which color, which breed, etc.

And around 1 million images were given bounding box labeling. Bounding Box means suppose we took an image of apple so it will show were the exact object is in the image, which will help during the task of object localization. So this is ImageNET classification with deep convolutional neural networks.

Now let’s talk about “How”, How he made all this possible?

So they used crowd sourcing meaning they took help of local people and ask them what do you see in these images using a service from Amazon called “Amazon Mechanical Turk” to build this entire database. And this is the dataset which changed the future of Deep Learning with CNN.

Let us discuss how the future of Deep Learning changed.

To know how the future of deep learning changed let’s discuss the next topic which is ImageNET Challenge.

As the big dataset was built, then the researchers decided to use this big dataset to start a competition, which is called ILSVRC – ImageNET Large Scale Visual Recognition Challenge simply put ImageNET Challenge.

So this challenge was started in 2010 and its goal was to bring up the best image classification models which most of the people should use. And a lot of people participated in this competition and the dataset which was used for this was the subset of the original image dataset, so in the original dataset there are 140 million images and dataset which was used for this had around 1 million images and only had 1000 categories. And in this way they reduced the complexity of the competition.

So a lot of teams or researchers participated in this competition in 2010, but in the first 2 years the models which came was of machine learning based models were deep learning was not used. These were machine learning based models so people where manually performing the feature extraction.

And the team which won the challenge in 2010 using machine learning model, their error rate was around 28%. 28% error rate means, if we show 100 images to this model it will give 28 wrong answers.

Also in 2011 again a machine learning model won the challenge and the error rate was improved a little and it was around 25%.

A featured change was found in 2012 and the year 2012 is also known as a landmark year for Deep Learning.

Because in the year 2012 Geoffrey Hinton participated with his CNN model named AlexNet which was totally based on deep learning. This was the first time CNN model was participated and he used GPU which boosted the training process. And this was also the first time when a CNN model used reLU as an activation layer.

As Geoffrey Hinton bought his CNN model AlexNet it made a revolution and the error rate was inclined to 16% from 25%.

Table of Evolution of models

YearModelError Rate
2010Ml model28%
2011Ml model25%
2012Alex NET16.4%
2013ZF NET11.7%
2014VGG7.3%
2015Google NET6.7%
2016Res NET3.5%
Evolution from 2010 – 2016

And in this way the concept of Deep Learning and CNN model came across the globe and because of deep learning pre trained models came into existence.

More Discussion about Pre Trained Models

Pre trained model

Pre trained models like Chat GPT – Chat Generative Pre Trained Transformer has change the whole sector of artificial intelligence technology. These models help us to get ahead in different tasks, which include image classification. TensorFlow which is an open source software library provides many different types of pre trained models which can be easily used for image classification and other AI tasks. Visit here to know more – https://www.tensorflow.org/tutorials/images/classification

Deep Learning which is a subset of Machine Learning has changed artificial intelligence to next level because of automatic extraction of difficult features. And because of Python Deep Learning and its training process of AI models have evolved much better.

Pre-trained models have made an evolution by using deep learning and particularly in the field of computer vision with Convolutional Neural Networks (CNNs). And they will evolve more in the coming times.

FAQs

What is Deep Learning?

Deep Learning comes from Machine Learning which is a part of machine learning that focuses on training itself. It tries to acts like a human brain were it tries to learn and extract hard patterns from large amount of data.

What is CNN convolutional Neural Network?

It a deep learning model mostly used for processing data like images. Nowadays CNN is most widely used for image classification purposes. It requires labeled visual (image) data.

What is Neural Network?

Neural Network in computers is inspired by human brain’s structure and functioning. It also tries to do the same functions as a human brain like understanding and training a set of data all by itself. Neural Networks has evolved the field of artificial intelligence to an advanced level which helps us to solve problems way faster than human brain.

What is Pre trained models?

Pre trained models are those models which are trained using large sets of data by some experts or researchers which is totally based on machine learning and deep learning. We can directly import these models into our projects easily. It performs tasks like object detection, image classification, etc.

Links to visit:

To know more about AI – https://innovatequests.com/categories/

To know what is chat generative pre-trained transformer – https://innovatequests.com/chat-gpt-ai-powered-chat-bot-2023/

Want to explore AI art generator – https://innovatequests.com/top-free-ai-art-generator-2023/

Know What is the exact definition and importance of AI – https://innovatequests.com/artificial-intelligence-explained-2023-cost-free-knowledge/

8 Comments on “Pre Trained Models and Deep Learning with CNN: Know the enhancement of AI 2023”

Leave a Reply

Your email address will not be published. Required fields are marked *