Lecture Note
Convolutional Neural Networks (CNN) Convolutional Neural Networks, or CNNs for short, are a subset of artificial neuralnetworks that were created with the particular purpose of processing and analysingstructured grid-like input, including photos and movies. CNNs have had a significantimpact on artificial intelligence, especially when it comes to computer vision tasks. The following are the main elements and ideas connected to CNNs: Layer of Convolution: This is the foundational element of a CNN. The input data aresubjected to convolutional filters, also referred to as kernels. These filters aid in therecognition of several elements in photos, including edges, textures, and patterns. Pooling Layer: Downsampling is used in pooling layers to minimise the spatialdimensions of the data. The most popular techniques to minimise the dimensionswhile keeping the most crucial information are max pooling and average pooling. Activation Functions: By introducing non-linearity to the network, activationfunctions allow the network to learn intricate correlations within the data. ReLU(Rectified Linear Activation), sigmoid, and tanh are examples of common activationfunctions. Fully Connected Layer: Also referred to as a dense layer, this layer links every neuronin the layer below to every neuron in the layer above. It is applied to finalclassifications or forecasts. Convolution: The process of convolution is swiping a filter over the input data,element-wise multiplying the filter and the input data, then adding the results. Thisprocedure supports feature extraction. Stride: The convolutional filter goes across the input data in steps of a certain sizecalled a stride. It establishes the amount of the filter's shift throughout each operation. Before applying convolution, padding entails adding additional rows and columns ofzeros to the input data. In order to preserve information at the borders of a picture, itmight be helpful to keep the spatial dimensions after convolution. A tiny matrix called the kernel, also known as the filter, is convolved with the inputdata. Edges, corners, and textures are examples of characteristics that various kernelsmay recognise. In a variety of AI tasks, CNNs have demonstrated extraordinary success, including: In the renowned ImageNet Large Scale Visual Recognition Challenge, CNNs wereemployed to achieve cutting-edge performance in categorising objects within pictures.
Object detection: CNNs are capable of locating and identifying various things insidean image, which is crucial for operations like surveillance and autonomous driving. Semantic segmentation allows for a more in-depth comprehension of an image'scontent since CNNs may divide an image up into many parts and give each region aname. Face Recognition: CNNs have been used to extract and recognise facial features,advancing biometric identification technologies. CNNs are used to analyse medical pictures like X-rays and MRIs, which helps in thediagnosis and identification of diseases. The capacity of CNNs to automatically learn hierarchical feature representations hasmade them an effective tool for managing complicated and high-dimensional data,such as photos and movies. They have been crucial in developing AI, particularly incomputer vision-related applications.
Unveiling Convolutional Neural Networks: Powering AI Vision
Please or to post comments