Chapter_2 Generative AI Introduction and Applications.pdf

Chapter 2: Generative AI Introduction and Applications Dr. Labed Abdeldjalil September 01, 2024 An Introduction To Generative AI Introduction: 1-2 Artificial Intelligence Simulation of human intelligence by machines Introdu...

Chapter 2: Generative AI Introduction and Applications Dr. Labed Abdeldjalil September 01, 2024 An Introduction To Generative AI Introduction: 1-2 Artificial Intelligence Simulation of human intelligence by machines Introduction: 1-3  Artificial Intelligence AI models Learn from massive existing data through training There are Two fundamental approaches to AI Discriminative AI Generative AI Introduction: 1-4 Discriminative AI ▪ Distinguishes between different ▪ Use advanced algorithms to: classes of data Differentiate ▪ Each data point labelled with its Classify class Identify patterns Draw conclusion ▪ Example: Email spam filters Introduction: 1-5 Discriminative AI ▪ Distinguishes between different ▪ Limitations: classes of data Cannot understand context Cannot generate new content Introduction: 1-6 Generative AI ▪ Create new content based on training data Capture the underlying distribution Generate novel data points Introduction: 1-7 Generative AI Prompt Output Text Text Image Image Video Audio Other form of input Video Code Data Text Text Text Image Image Video Introduction: 1-8 Discriminative AI versus Generative AI “AI can not only boost our analytic and decision-making abilities but also heighten creativity.” -Harvard Business Review Introduction: 1-9 Deep learning and neural networks Train artificial neural networks on massive data Introduction: 1-10 Evolution of generative AI Introduction: 1-11  Foundation Models AI models with broad capabilities Adapted to build specialized and Large Language models: Process advanced models or tools and generate text A specific category of foundation models called (LLMs) Introduction: 1-12  Foundation Models Examples ▪ Example of LLMS OpenAI: GPT n-series (GPT 1, GPT 2, GPT-3/3.5 and GPT4) Google: Gemini Meta: Llama ▪ Example of models for image generation Stable Diffusion DALL-E Midjourney Introduction: 1-13 Generative AI tools Generative AI for diverse use cases ChatGPT Text generation Gemini DALL-E Image generation Midjourney Video generation Synthesia Copilot Code generation AlphaCode Gemini Introduction: 1-14 Capabilities of Generative AI Introduction: 1-15 Capabilities of Generative AI Introduction: 1-16 Text generation capabilities of generative AI Large Language models (LLMs) Trained on large data sets Generate human-like text Introduction: 1-17 Text generation capabilities of generative AI Large Language models (LLMs) Learn patterns and structure from dataset Generate coherent and contextually relevant Text or response Conversation Explanation Summaries Introduction: 1-18 Text generation capabilities of generative AI Introduction: 1-19 Text generation capabilities of generative AI Introduction: 1-20 Image generation capabilities of generative AI ▪ Generative AI models leverage deep learning techniques Generative adversarial networks (GANs) Variational autoencoders (VAEs) ▪ Generative AI images Realistic textures Natural colors Fine-grained details Introduction: 1-21 Image generation capabilities of generative AI ▪ Image generation generative AI models StyleGAN High-quality High-resolution Novel images DeepArt Complex and detailed artwork from a sketch DALL-E Novel images based on textual description Introduction: 1-22 Image generation capabilities of generative AI Introduction: 1-23 Audio generation capabilities of generative AI Introduction: 1-24 Audio generation capabilities of generative AI Introduction: 1-25 Audio generation capabilities of generative AI WaveGAN Raw audio waveforms Realistic sound (speech, music) OpenAI’s MueNet Original music in various genres and instrumentations Classic composition to pop song Mozilla TTs and Google’s Tacotron 2 High realistic synthetic speech (Tone, pitch, Rhythm, Expression) Introduction: 1-26 Audio generation capabilities of generative AI Introduction: 1-27 Video generation capabilities of generative AI Introduction: 1-28 Video generation capabilities of generative AI Introduction: 1-29 Video generation capabilities of generative AI Introduction: 1-30 Video generation capabilities of generative AI Introduction: 1-31 Video generation capabilities of generative AI Introduction: 1-32 Video generation capabilities of generative AI Introduction: 1-33 Video generation capabilities of generative AI Introduction: 1-34 Code generation capabilities of generative AI Introduction: 1-35 Code generation capabilities of generative AI Introduction: 1-36  Code generation capabilities of generative AI GitHub Copilot AI-based programming assistants Can generate code for various programming language and frameworks Powered by OpenAI Codex Autocomplete code Accelerate hard tasks Generate code per input Can integrate with code editor Introduction: 1-37 Code generation capabilities of generative AI Introduction: 1-38 Virtual world creation generation capabilities of generative AI Virtual avatars simulating Realistic behavior Expression Conversation Decisions Introduction: 1-39 Virtual world creation generation capabilities of generative AI ▪ Complex virtual environments Realistic textures Sound Objects Personalized expression Virtual identities with unique personalities Introduction: 1-40 Virtual world creation generation capabilities of generative AI Gaming Entertainment Education Augmented and virtual reality Metaverse platforms Virtual influencers and digital personalities Introduction: 1-41