7: Deep Learning and AI Basics
( \newcommand{\kernel}{\mathrm{null}\,}\)
- 7.0: Introduction
- This page discusses the importance of neural networks in AI, highlighting their structure's similarity to the human brain for pattern recognition and prediction. It covers how neuroscience informs our understanding of neuron connections. While early AI faced computational power limitations, advancements have led to deep learning models that empower complex applications like natural language processing. Additionally, the text tackles ethical concerns associated with AI and machine learning.
- 7.1: Introduction to Neural Networks
- This page covers the fundamentals of neural networks, including their structure, essential components, and applications in image recognition and speech processing. It introduces key concepts like weights, biases, activation functions, and the multilayer perceptron (MLP). The text explains various activation functions and the perceptron learning rule, illustrated through a case study using the Iris dataset.
- 7.2: Backpropagation
- This page discusses the fundamentals of neural networks, focusing on weight and bias adjustments through backpropagation to minimize errors in supervised learning. It covers loss functions, gradient descent, and the importance of differentiability in optimizing models. The implementation of a neural network on the MNIST dataset using TensorFlow is described, alongside the architecture and challenges of recurrent neural networks (RNNs), particularly long short-term memory (LSTM) networks.
- 7.3: Introduction to Deep Learning
- This page provides an overview of deep learning principles, focusing on neural networks and the role of hidden layers in feature recognition and classification. It discusses loss functions, including mean squared error and binary cross entropy, crucial for training. The text highlights the use of sparse categorical cross entropy with softmax for multi-class tasks, demonstrating its application in classifying handwritten numerals using TensorFlow.
- 7.4: Convolutional Neural Networks
- This page offers an overview of convolutional neural networks (CNNs) and their effectiveness in image processing tasks such as classification, object detection, and semantic segmentation. It describes CNN components, including convolutional, pooling, and fully connected layers, and explains feature maps that capture hierarchical image features.
- 7.5: Natural Language Processing
- This page discusses the evolution of Natural Language Processing (NLP) from basic systems to sophisticated models like ChatGPT, highlighting key advancements and applications in various fields. It notes the transformative role of AI in creative industries and workplace productivity, alongside challenges such as ethical concerns regarding intellectual property and the impact on human creativity.
- 7.6: Key Terms
- This page serves as a glossary of key terminology and definitions related to neural networks and artificial intelligence, covering concepts such as neuron activation, activation functions, deep learning, CNNs, RNNs, backpropagation, gradient problems, loss functions, and the significance of different layers. It highlights various AI applications, including natural language processing and generative art, making it a comprehensive resource for foundational AI and machine learning knowledge.
- 7.7: Group Project
- This page discusses three machine learning projects: Project A aims to develop a neural network for diagnosing cirrhosis, focusing on data preparation and model training with TensorFlow. Project B involves building a convolutional neural network to identify handwritten digits from the MNIST dataset, promoting experimentation. Project C uses natural language processing and AI art for creating an illustrated story, highlighting theme development, narrative coherence, and visual-text alignment.
- 7.8: Chapter Review
- This page discusses the role of hidden layers in neural networks for learning complex patterns, highlights the effectiveness of convolutional neural networks in image-related tasks, and emphasizes advancements in speech recognition for improved user experience. It also addresses ethical concerns regarding AI, particularly the unauthorized data collection by virtual assistants.
- 7.9: Critical Thinking
- This page discusses neural network models, starting with a single perceptron for flu diagnosis using specific markers and weight adjustment. It evaluates standard, recurrent, and convolutional networks for tasks such as sentiment analysis, stock prediction, image classification, and subscription cancellation. It also explores bitmap character representations and image reduction techniques like pooling, concluding with the identification of suitable loss functions for diverse predictive tasks.
- 7.10: Quantitative Problems
- This page describes a neural network designed to classify students' likelihood of graduating using various activation functions (ReLU, Leaky ReLU, Sigmoid, Softplus). It includes an analysis of the most effective activation function for classification and the method for determining the output. The page also discusses calculating average loss using Mean Squared Error (MSE) and Hinge loss for predictions on different data points.