Scribd
Upload
3 views3 pages

Nueral Network

The document discusses the fundamentals of Convolutional Neural Networks (CNN) and their role in artificial intelligence and computer vision. It outlines the differences between AI and robots, the steps involved in CNN, and the significance of data augmentation and K-fold cross-validation. Additionally, it touches on pre-trained networks, confusion matrices, and the importance of dataset selection for training models.

Uploaded by

Sh Rahat
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
3 views3 pages

Nueral Network

The document discusses the fundamentals of Convolutional Neural Networks (CNN) and their role in artificial intelligence and computer vision. It outlines the differences between AI and robots, the steps involved in CNN, and the significance of data augmentation and K-fold cross-validation. Additionally, it touches on pre-trained networks, confusion matrices, and the importance of dataset selection for training models.

Uploaded by

Sh Rahat
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

CNN Class

Dr. Md Saiful Islam


Date : 29 Jan,2025
Artificial- Not Natural Intelligence, Those who have no brain.The machines. They’ve done the work
earlier ruled base like Robot, Specific Tasks.
Intelligence- What is intelligence?
Differences between AI and Robot.
AI Robot
Programmed to think Programmed to do
Social Interaction Low-level interaction
Learns Only as smart as a program
How do we make Machienes Think?
By using the neural Network ,Extracting the Features of the Images.Computer Understands only the 1/0.
So features are converted to these 1/0’s.
Training the AI via Pattern using Algorithm. Like Gradient Decent(Picks the Lowest point marvel
falling).Accuracy 100% can be achieved using Gradient decent. Today Every Algorithm has behind
scenario of Gradient Decent Algorithm.
Challenges:
1. To Extract the features from Images/text.While the symptomps are so close.To address the
Challenge Covolution Neural network is Introduced (CNN).CNN extract the feature

Neural Networks:
Neural Networks has 1000 has stride.There are different types of neural network. But we’ll study the
Convolution Neural Network (CNN).CNN is the Foundation of Computer Vision.

Image Output Output


Accuracy of features Depends on size,More size more Accurate.
CNN Steps Basic 2 Steps, Can be further divided into 4 more steps.

1. Convolution Layer[There may be an Activation Layer ReLU after Convolution Layer that’s nulyfy the
Negative values.]
2. Pooling Layer
3. Flattening
4. Full Connection.
How does CNN work?
Convolution Layer:
CNN is a Feature Detector that takes the O/P data randomly (1/0), And does the Convolution step by step
and Creates a Featured map.
More Featured Detector creates a more featured map.
Then the features are extracted using some kind of filtering. Deep learning means using more
Convolution Layers.
Pooling layer:
Reduces the Size using Max pooling or Avg Pooling,
Flattening:
To represent a Matrix in a Column from a pooled featured map. Then we get the Features.
Full Connection:
Connected to the Neural Network. There are Thousands of Classifiers SVM, KNN,Neural Networks etc
etc.
Now lets Do it in Python.
https://colab.research.google.com/drive/1oUiE3Q-
fNjB2wqm82EPt14gJsyPnHZgD#scrollTo=lmQ76vQKDf9s
2. https://keras.io/examples/vision/image_classification_from_scratch/

Image Data Augmentatoion:


To create Data Artificially,By shifting images Horizontally and Vertically.
Significance of Augmentation?
How to do Augmentation in Tabled data?
Using Interpolation method.
K-fold Cross Validation:[Before Data]
Total Data sets are Divided into 2 parts, namely Training (80%) and Validation (20%).Training
and Validation Accuracy defines the Total accuracy by Averaging.
Training and Validation data must be close to each other To optimize the Accuracy.
To divide the data closely the data are divided using K-fold Validation.

Pre-trained Network / Transfer Learning:


Already Trained Models, That are trained with Millions of Data.If this trained data is used then
we call it to Transfer Learning.
Advantages and Disadvantages of the Pretrained Model.
Some Pre-trained models are, VGG, Google Net, etc.
What are the Confusion Matrix’s significance?
What is F1 Score, Recal Score, and Specificity?
Data sets are available at Kaggle.com, Same work can be done using Kaggle Notebook,
Kaggle Notebook Is faster than Colab.
Is data set used to any Journal?
Satelite image, Wafer Dataset.Choose Data set particular to you can go through Long run.
Data set can be, Image,Text,Video etc etc.

Next Task
To Read some Research Papers from Google Scholar and so on.

You might also like