Convolutional neural networks
Download as PPTX, PDF0 likes105 views
Convolutional neural networks apply convolutional layers and pooling layers to process input images and extract features, followed by fully connected layers to classify images. Convolutional layers convolve the image with learnable filters to detect patterns like edges or shapes, while pooling layers reduce the spatial size to reduce parameters. The extracted features are then flattened and passed through fully connected layers like a regular neural network to perform classification with a softmax output layer. Dropout regularization is commonly used to prevent overfitting.
Convolutional neural networks
- 2. INTRODUCTION ā¢ Tensor flow is an open source machine learning library for research and production. ā¢ Tensor means having both magnitude and direction. ā¢ Tensor flow can be represented in n-dimension.
- 3. Neural Networks ā¢ Neural Networks which are inspired from biological neurons. ā¢ Neural networks which has the following functions ā¢ Neurons ā¢ Connection Links ā¢ Weights ā¢ Activation Functions ā¢ Convolutional Neural Network= Convolution + Neural Network
- 5. Convolution Layer ā¢ Convolution is the first layer to extract features from an input image. ā¢ Convolution preserves the relationship between pixels by learning image features using small squares of input data. ā¢ Convolution of an image with different filters can perform operations such as edge detection, blur and sharpen by applying filters ā¢ It is a mathematical operation that takes two inputs such as image matrix and a filter or kernel
- 6. ā¢ Here if input is 50x50 matrix is multiplied with stride=2 and filter of 3x3 matrix we get an output which is called āfeature mapā of size 25x25.
- 7. Strides ā¢ Stride is the number of pixels shifts over the input matrix. ā¢ When the stride is 1 then we move the filters to 1 pixel at a time. When the stride is 2 then we move the filters to 2 pixels at a time and so on
- 8. Padding ā¢ Sometimes filter does not fit perfectly fit the input image Then Pad the picture with zeros (zero-padding) so that it fits
- 9. Rectified Linear Unit (RELU) ā¢ The main purpose of Relu is to introduce non-linearity ā¢ The ReLU maps input x to max (0, x), that is, they map negative inputs to 0, and positive inputs are output without any change as shown.
- 10. Pooling Layer ā¢ Pooling layer reduce the number of parameters when the images are too large. ā¢ Different types of pooling: ā¢ Max Pooling ā¢ Average Pooling ā¢ Sum Pooling ā¢ Max pooling is widely used one as it take the largest element from the rectified feature map.
- 11. Fully Connected: ā¢ The layer is Fully Connected layer, we flattened our matrix into vector and feed it into a fully connected layer like neural network. ā¢ With the fully connected layers, we combined these features together to create a model.
- 12. Softmax ā¢ Finally, we have an activation function such as softmax or sigmoid to classify the outputs. ā¢ Softmax Converts outputs to probabilities by dividing output by summation of all the values ā¢ Forces the output be sum of 1
- 13. Cross-Entropy ā¢ Cross-entropy is a loss function for which error has to be minimized. ā¢ Compares the distance between output of softmax and one hot encoding.
- 14. Dropout ā¢ Dropout is an effective way of regularizing neural networks to avoid the overfitting ā¢ Drop out should be above 0.3
- 15. Result ā¢ The performance was evaluated by using accuracy of the model with actual vs predicted. ā¢ How to find out best number of hidden layers for a model? ā¢ If image size is nxn ā¢ stride-=s ā¢ Then no of hidden layers L should be (n/s)=a ā¢ i.e., number of layers L should always be greater than a.