Essentials of Generative Artificial Intelligence PDF

Essentials of Generative Artificial Intelligence Unit 1  1. Introduction Generative Modeling Generative AI refers to a class of artificial intelligence systems designed to create new content, such as text, images, audio, videos, code, or othe...

Essentials of Generative Artificial Intelligence Unit 1  1. Introduction Generative Modeling Generative AI refers to a class of artificial intelligence systems designed to create new content, such as text, images, audio, videos, code, or other types of data. https://www.youtube.com/watch?v=rwF-X5STYks https://www.youtube.com/watch?v=rwF-X5STYks  Image generator into the new MetaAI chatbot Meta AI to draw a photo of two dogs in karate uniforms holding nunchucks and this was the result (Image credit: Meta AI)  1. Introduction (cont..) Problem can be solved using generative modeling Suppose we have a dataset containing images of horses. We may wish to build a model that can generate a new image of a horse that has never existed but still looks real because the model has learned the general rules that govern the appearance of a horse. Generative Modeling  Example First, we require a dataset consisting of many examples of the entity we are trying to generate. This is known as the training data, and one such data point is called an observation. First, we require a dataset consisting of many examples of the entity we are trying to generate. This is known as the training data, and one such data point is called an observation. Generative AI with a text-guided image-editing tool.  1. Introduction Generative models create new images by learning patterns and features from a large database of photographs. The images generated are similar to the ones the model was trained on, but can also be completely new and original. This technique is known as image synthesis. Variety of generative AI models has its own advantages and disadvantages. Generative Modeling  2. Generative Adversarial Networks (GANs) GAN is a popular and effective type of generative AI model used for creating images. GAN consists of two neural networks: 1. Generator network 2. Discriminator network Generator network creates new images. Discriminator network determines if the images created by the generator are real or fake. Generative Modeling During the training process, the two networks are trained in parallel, in a technique known as adversarial training. The generator tries to trick the discriminator, while the discriminator tries to distinguish between real and fake images. As a result, the generator learns to create images that are increasingly realistic and difficult for the discriminator to identify as fake.  3. Variational Autoencoders (VAEs) Another type of generative AI model used for picture synthesis. VAEs are networks that consist of 1. Encoder 2. Decoder The encoder learns a compressed representation of an input image, also known as latent space. The decoder uses this compressed representation Generative Modeling to generate new images that are identical to the input image. When combined with other methods like adversarial training, VAEs have shown promising outcomes in creating high-quality images. They are capable of generating graphics with intricate features such as textures and patterns, and can manage complicated visuals. Additionally, the encoding and decoding processes used by VAEs have a probabilistic component, which enables them to produce a wide range of new pictures from a single input image.  3. Variational Autoencoders (VAEs) unlike GANs, VAEs may have difficulty in producing extremely realistic pictures. They also take longer to produce images since each new image needs to be encoded and decoded. Despite these drawbacks, VAEs continue to be a widely used method for image synthesis and have shown effectiveness in various applications such as computer graphics and medical imaging. Generative Modeling  4. Autoregressive models Autoregressive models are a type of generative AI model used for image creation, where the model starts with a seed image and creates new images pixel by pixel. The model predicts the value of the next pixel based on the values of the preceding pixels. While autoregressive models can create high-quality photos with intricate details, they produce new images relatively slowly because each pixel must be generated separately. Generative Modeling Despite this limitation, autoregressive models have demonstrated effectiveness in producing high-quality images with fine details and complex structures, particularly in applications such as picture inpainting and super-resolution. However, compared to GANs, autoregressive models may have difficulty in producing extremely realistic images  5. Choosing the right dataset for your model Generative AI models rely heavily on the dataset they are trained on to generate high-quality, diverse images. To achieve this, the dataset should be large enough to represent the richness and variety of the target picture domain, ensuring that the generative model can learn from a wide range of examples. For example, if the goal is to create medical images, the dataset should contain a diverse range of medical photos capturing Generative various illnesses, organs, and imaging Modeling modalities.  Vector Embedding Vector embeddings are mathematical representations of data that convert unstructured data into a format that machine learning models can process: Prompt engineering is the process of designing and optimizing inputs, or prompts, to guide AI models to generate desired outputs.  6. Generative Versus Discriminative Modeling Dataset of paintings, some painted by Van Gogh and some by other artists. Train a discriminative model to predict if a given painting was painted by Van Gogh. The model would learn that certain colors, shapes, and textures are more likely to indicate that a painting is by the Dutch master, and for paintings with these features, the model would upweight its prediction accordingly. Generative Modeling  6. Generative Versus Discriminative Modeling When performing discriminative modeling, each observation in the training data has a label. For a binary classification problem such as our artist discriminator, Van Gogh paintings would be labeled 1 and non–Van Gogh paintings labeled 0. The model then learns how to discriminate between these two groups and outputs the probability that a new observation has label 1—i.e., that it was painted by Van Gogh. Generative Modeling  6. Generative Versus Discriminative Modeling Discriminative modeling learning a function that maps an input to an output using a labeled dataset. (form of supervised learning) Generative modeling is usually performed with an unlabeled dataset (form of unsupervised learning), though it can also be applied to a labeled dataset to learn how to generate observations from each distinct class. Mathematical notation to describe the difference between generative and discriminative modeling. Generative Modeling  6. Generative Versus Discriminative Modeling Discriminative modeling attempts to estimate the probability that an observation x belongs to category y. Generative modeling doesn’t care about labeling observations. The key point is that even if we were able to build a perfect discriminative model to identify Van Gogh paintings, it would still have no idea how to create a painting that looks like a Van Gogh. It can only output probabilitiesGenerative against existing Modeling images, as this is what it has been trained to do. We would instead need to train a generative model, which can output sets of pixels that have a high chance of belonging to the original training dataset.  7. Advances in Machine Learning For example, for image classification, the key breakthrough came in 2012 when a team led by Geoff Hinton at the University of Toronto won the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) with a deep convolutional neural network. The competition involves classifying images into one of a thousand categories and is used as a benchmark to compare the latest state-of-the-art techniques. The deep learning model had an error rate of 16%—a massive improvement on the next best model, which only achieved a 26.2% Generative error rate. Modeling This sparked a deep learning boom that has resulted in the error rate falling even further year after year. The 2015 winner achieved superhuman performance for the first time, with an error rate of 4%, and the current state-of-the-art model achieves an error rate of just 2%. Many would now consider the challenge a solved problem.  8. The Rise of Generative Modeling While exciting, this also raises ethical questions around the proliferation of fake content on the internet and means it may become ever harder to trust what we see and read through public channels of communication. Generative Modeling The Generative Modeling Framework Generative Modeling  The Generative Modeling Framework While this isn’t the most complex example, we can use it to understand what generative modeling is trying to achieve. The following framework sets out our motivations. Generative Modeling The Generative Modeling Framework Modeling  The Generative Modeling Framework As we can see from Figure 1-6, the data-generating rule is simply a uniform distribution over the land mass of the world, with no chance of finding a point in the sea. Modeling  The Generative Modeling Framework Modeling The field of generative modeling is diverse and the problem definition can take a great variety of forms. However, in most scenarios the Generative Modeling Frame-work captures how we should broadly think about tackling the problem.  Probabilistic Generative Models As a first step, we shall define four key terms: sample space, density function, parametric modeling, and maximum likelihood estimation. eg Probabilistic Generative Models Modeling Probabilistic Generative Models Modeling Probabilistic Generative Models Modeling Probabilistic Generative Models Modeling Probabilistic Generative Models ling  Probabilistic Generative Models A dataset featuring 50 observations of Wrodler fashion (Figure 1-7) and told that you have a day to come up with 10 new styles to present to the Fashion Police for inspection. You’re allowed to play around with hair styles, hair color, glasses, clothing type, and clothing color to create your masterpieces. ling  As you’re a data scientist at heart, you decide to deploy a generative model to solve the problem.  Your First Probabilistic Generative Model Wrodl dataset consists of N = 50 observations of fashions currently seen on the planet. Each observation can be described by five features, (accessoriesType, clothingColor, clothingType, hairColor, topType), as shown in Table 1-1. ling Your First Probabilistic Generative Model ling Practicals: https://www.youtube.com/watch?v=lV82LDKT16A THANK YOU