Scribd
Upload
2 views

deep learning u3

Uploaded by

shivamchoubeyrishu4747
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
2 views

deep learning u3

Uploaded by

shivamchoubeyrishu4747
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

1.

Introduction to Convolutional Neural Networks (CNN)

• Definition: CNNs are a class of deep learning neural networks commonly used for
analyzing visual imagery. They are specifically designed to process pixel data and
recognize patterns.

• Architecture: Typically composed of a series of layers, including convolutional layers,


pooling layers, and fully connected layers.

• Use Cases: Image classification, object detection, facial recognition, and more.

2. CNN Terminologies

• ReLU (Rectified Linear Unit) Activation Function: An activation function defined as


f(x)=max⁡(0,x)f(x) = \max(0, x). It introduces non-linearity to the model and helps to
accelerate convergence.

• Stride: Refers to the number of pixels by which the filter slides over the input feature
map. A larger stride reduces the spatial dimensions of the output.

• Padding: Adding extra pixels around the border of the input feature map to control the
spatial dimensions of the output. Common types are 'valid' (no padding) and 'same'
(padding to keep output dimensions same as input).

• Pooling: A down-sampling operation that reduces the dimensionality of the feature


map. Common types are max pooling and average pooling.

• Convolutions Operations: The process of applying a filter (kernel) to an input feature


map to produce a transformed feature map, capturing spatial hierarchies.

• Convolutional Kernels: Small, learnable filters that slide over the input to detect
features like edges, textures, and patterns.

3. Types of Layers

• Convolutional Layer: Applies convolutional operations to the input, producing a


feature map.

• Pooling Layer: Reduces the spatial dimensions of the feature map, usually after a
convolutional layer.

• Fully Connected Layer: Connects every neuron in one layer to every neuron in the next
layer, used at the end of the network to make predictions.

4. Visualizing CNN
• Tools and techniques like saliency maps, gradient-weighted class activation mapping
(Grad-CAM), and activation maximization are used to understand what CNNs learn and
how they make decisions.

5. CNN Examples

• LeNet: One of the earliest CNN architectures, primarily used for handwritten digit
recognition.

• AlexNet: Introduced deeper layers and ReLU activation, won the ImageNet challenge in
2012.

• ZF-Net: An extension of AlexNet, with improvements in filter sizes and number of


layers.

• VGGNet: Known for its deep but simple architecture with 16-19 layers, using very small
(3x3) convolution filters.

• GoogLeNet (Inception): Introduced the Inception module that uses multiple filter sizes
to capture different features at various scales.

• ResNet: Introduced residual connections to enable very deep networks by addressing


the vanishing gradient problem.

• RCNN: A series of networks for object detection that combine region proposals with
CNNs for accurate detection.

6. Deep Dream

• A technique using CNNs to modify an image to enhance certain features, creating


dream-like effects. It helps to visualize what the network has learned.

7. Deep Art

• Uses neural networks to apply artistic styles to images, creating artwork that mimics
famous styles (e.g., Van Gogh, Picasso).

8. Regularization Techniques

• Dropout: Randomly sets a fraction of input units to zero during training to prevent
overfitting.

• Drop Connect: Similar to Dropout but drops connections between neurons rather than
the neurons themselves.

• Unit Pruning: Removes unimportant neurons to reduce the complexity of the model.
• Stochastic Pooling: Uses a probabilistic approach to pooling, selecting values based on
their magnitude.

• Artificial Data (Data Augmentation): Generates new data samples by applying


transformations like rotation, scaling, and cropping to the existing dataset.

• Injecting Noise in Input: Adds random noise to input data to make the model more
robust.

• Early Stopping: Stops training when the model's performance on a validation set starts
to degrade.

• Limit Number of Parameters: Reduces the model size to prevent overfitting.

• Weight Decay: Adds a penalty term to the loss function proportional to the magnitude
of the weights to prevent large weights.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy