Artificial Neural Networks (ANN)

Introduction and Update

• Depleted Playlist Update: New videos in the playlist.
• Topics Covered:
  • Basic Concepts
  • How to Improve Neural Networks
  • Some Topics Left (e.g., Colgate and Cancer)

New Topic: Convolutional Neural Networks (CNN)

• Introduction:
  • CNN = Convolutional Neural Networks
  • Use: Image and Time Series Data Processing
  • Applications: Facial Recognition, Self-Driving Cars

Basic Components of CNN

• Convolutional Layers: Feature Extraction
• Pooling Layers: Data Size Reduction
• Fully Connected Layers: Every node is connected to each subsequent node
• Special Operations: Convolutional Operations (different from matrix multiplication)

Inspiration and History

• Inspiration: Human Brain (Visual Cortex)
• History: 1998, Contribution of Yann LeCun
  • First successful CNN
  • Microsoft’s OCR, Handwriting Recognition Tools

Difference Between ANN and CNN

• Operationally: Wavelet Transform and Favorite Enhancement in CNN
• Feature Extraction: In Initial Layers
• Classification: In Later Layers
• Confusion Matrix: Measurement Point
• Computational Cost: Increases Based on Data Size

Need and Advantages

• Why CNN?: Better Performance on Image Data, especially with regard to pixel arrangement
• Limitations of ANN:
  • High Computational Cost
  • Overfitting Problems
  • Improper Feature Capture

How CNN Works

• Convolutional Layers: Basic Feature Extraction
• Primitive Features: Lines, Edges, etc.
• Complex Features: Aggregated Later
• Image Classification Task: Example of Digit ‘9’

Applications

• Primary: Image Classification, Object Detection, Face Detection, Image Segmentation
• Advanced: Resolution Enhancement, Black and White to Color Images
• Real Time: Sports Pose Detection, Smartphone Camera, Self-Driving Cars

Future Plans

• Curriculum: Basics, Biological Connection, Operations (Convolution, Stride, Padding), C&A Examples, Transfer Learning

I hope these notes give you a clear and useful understanding of CNN.