Al: Introduction to Computer Vision

In the previous article in this series on Al and machine learning, we explored natural language processing (NLP) using the packages PyTorch and NLTK. We also discussed PySpark and how it can be used in Big Data analytics. The main objective of this tenth article in the series on Al is to concentrate on computer vision.

- Deepu Benson

By now, we have become acquainted with various types of neural networks, including FNN, CNN, RNN, LSTM, etc, among others. However, to develop AI-based applications effectively, it is essential to gain knowledge about the different types of neural network architectures. It is also important to note that the terms ‘neural network models’ and ‘neural network architectures’ are often used interchangeably, but there is a subtle difference between them. A neural network architecture is concerned with the overall design of the network, which includes the number of layers, the types of layers (fully connected, convolutional, recurrent, etc), the activation functions, and the connectivity pattern between the layers. For instance, a neural network architecture may consist of a fully connected feedforward neural network with ten hidden layers and an ReLu activation function. Many neural network models can instantiate this architecture, each with its own set of parameter values learned from the training data for a specific task.