DS Unit 4 Sprint 15: Major Neural Network Architectures
Welcome to Sprint 15
Now that you've learned the foundations of Neural Networks, it's time to go deep! Of course, all "deep learning" really means is "there's at least some hidden layers" - but there's a great deal of variety both in the layered architecture and the behavior of individual "neurons" in the network.
We'll study a few of the most effective recent innovations in neural networks and deep learning and think a bit about what the future may hold. Is deep learning the path to artificial general intelligence? Probably not - but LLMs might get us part way there!
Sprint Modules
This sprint explores specialized neural network architectures that power today's most advanced AI applications in computer vision, natural language processing, and time series forecasting:
Module 1
Recurrent Neural Networks and LSTM
Traditional neural networks are feedforward, but recurrent neural networks introduce cycles that act as memory. This makes them ideal for sequential data like natural language, time series, and other contexts where order matters. Learn how LSTMs overcome the vanishing gradient problem.
View ModuleModule 2
Convolutional Neural Networks
CNNs draw from biological inspiration, using neuron connectivity patterns that resemble the brain's visual fields. These models excel at image recognition and classification, often achieving human-level performance with minimal preprocessing.
View ModuleModule 3
OpenAI and ChatGPT
Explore ChatGPT and transformer-based models that demonstrate AI's ability to understand and generate human-like text. Learn about the architecture, applications, and ethical considerations of these powerful language models.
View ModuleModule 4
Large Language Models
Work with LLMs through the OpenAI API and SDK for text generation and summarization. Learn to customize outputs using parameters and set up local LLMs for specialized applications like chatbots.
View Module