3.0 Deep Learning and Neural Networks Big Picture скачать в хорошем качестве

3.0 Deep Learning and Neural Networks Big Picture

The link between the inputs and the outputs of an AI system is often a mystery, operating within a black box. This is particularly true for artificial neural networks, or ANNs, and it raises some challenging ethical quandaries. This module explores some of the uses of neural networks and their role in deep learning—essential knowledge for any aspiring data scientist. We'll begin with the fundamentals of ANNs and their applications in various fields, quickly moving on to an exploration of their structure, focusing on the concept of layers. We'll cover input, hidden, and output layers, and explain the difference between shallow and deep networks, the building blocks of neural network architectures. We'll introduce a range of neural network architectures, including perceptrons, used in simple tools such as spam detection, and multilayered networks, essential for complex tasks such as image recognition. We'll also briefly cover modular networks, used in robotics, and liquid state machines, suited for time-series data such as step-detection. Finally, we'll look at transformers, including advanced architectures such as BERT and GPT, used in natural-language processing. To bring these concepts to life, we'll introduce the MNIST dataset used for handwritten digit recognition. You'll have the opportunity to build a simple ANN and see how it performs. We'll then explain how neural networks work, as data flows through the network from input to output—referred to as "feed forward"—and describe the learning process in which the network adjusts to minimize errors—referred to as "backpropagation". Training a neural network involves initialization, training, validation, and testing. We'll discuss the importance of each phase and why the use of separate datasets ensures that the model created in the network generalizes well. By the end of this module, you'll have a comprehensive understanding of neural networks and deep learning. You'll be ready to build and train your own neural networks and understand their application to a variety of fields.