Major types of generative AI back-end technologies and how they are used
There are various types of generative AI, each with different back-end technologies that have different characteristics and applications. This section briefly describes the back-end technologies of generative AI.
Posted at: 2023.5.16
Summary of main generative AI backend technology types and applications
| Types of Generative AI | Characteristics | Examples of optimal applications |
|---|---|---|
| GAN (Generative Adversarial Network) | Generates data by pitting two models, a generator and a discriminator, against each other. It is used to generate highly realistic images and sounds. | Automatic image generation, speech synthesis, and design generation |
| VAE (Variational Autoencoder) | Learns latent features of data and generates new data. Used to generate images, audio, text, etc. | Image generation, speech synthesis, anomaly detection |
| RNN (Recurrent Neural Network) | Generates the next step while remembering past information. Used for natural language processing and prediction of time series data. | Language modeling, sentence generation, music composition |
| Transformer | Generates sequence data using a self-attention mechanism. Used for natural language processing and image generation. | Automatic translation, sentence generation, image generation |
| StyleGAN (Style-based Generative Adversarial Network) | A type of GAN, specialized for high-resolution photorealistic image generation. | Photorealistic image generation, artwork creation |
| DALL-E (Differentiable Neural Network to Generate Innovative Images) | An AI model that generates images based on textual descriptions. Capable of detailed control of images and generation of unrealistic images. | Image generation from text, creative illustration production |
| DeepDream | Visualize the internal representation of neural networks and add fantastic visual effects to existing images. | Generation of art works, transformation and enhancement of images |
| MusicVAE | Learns latent features of music and generates new musical compositions. | Automatic composition, remixing of music |
| Sketch-RNN | Learns latent features of sketches, generates new sketches | Automatic sketch generation, design completion |
| Pix2Pix | Learns mappings between input images and corresponding output images, used for image style transformation and image restoration | Image transformation, image restoration, style transformation |
Back-end technology for sentence generation AI
Transformer is probably the most mainstream AI for sentence generation, and is used by "GPT" such as Chat GPT, Google Bard's "PaLM 2" and "BERT".
| Types of Generative AI | Characteristics | Examples of optimal applications |
|---|---|---|
| Transformer | Generates sequence data using a self-attention mechanism. Used for natural language processing and image generation. | Automatic translation, sentence generation, image generation |
| RNN (Recurrent Neural Network) | Generates the next step while remembering past information. Used for natural language processing and prediction of time series data. | Language modeling, sentence generation, music composition |
| GAN (Generative Adversarial Network) | Generates data by pitting two models, a generator and a discriminator, against each other. It is used to generate highly realistic images and sounds. | Automatic image generation, speech synthesis, and design generation |
Back-end technologies for image-generating AI
DALL-E (Differentiable Neural Network to Generate Innovative Images), GAN (Generative Adversarial Network), VAE (Variational Autoencoder) are the mainstream technologies.
| Types of Generative AI | Characteristics | Examples of optimal applications |
|---|---|---|
| GAN (Generative Adversarial Network) | Generates data by pitting two models, a generator and a discriminator, against each other. It is used to generate highly realistic images and sounds. | Automatic image generation, speech synthesis, and design generation |
| VAE (Variational Autoencoder) | Learns latent features of data and generates new data. Used to generate images, audio, text, etc. | Image generation, speech synthesis, anomaly detection |
| StyleGAN (Style-based Generative Adversarial Network) | A type of GAN, specialized for high-resolution photorealistic image generation. | Photorealistic image generation, artwork creation |
| DALL-E (Differentiable Neural Network to Generate Innovative Images) | An AI model that generates images based on textual descriptions. Capable of detailed control of images and generation of unrealistic images. | Image generation from text, creative illustration production |
| DeepDream | Visualize the internal representation of neural networks and add fantastic visual effects to existing images. | Generation of art works, transformation and enhancement of images |
| Sketch-RNN | Learns latent features of sketches, generates new sketches | Automatic sketch generation, design completion |
New Posts
- No ImageThis is all you need to remember about how AI works.
AI is a complex field with difficult words and complicated mechanisms, and in this article, we will explain at least this much about how AI works so that even beginners can understand it.
- No ImageHow to proofread text and check for errors with ChatGPT and its advantages and disadvantages
文Proofreading is a very time-consuming task, but by hiring ChatGPT, you can save a lot of stress, time, and most importantly, money.
- No ImageSummary of versatile phrases for ChatGPT & Bing Chat
Sometimes it is difficult to elicit a good response when using ChatGPT or Bing Chat. We have compiled a list of versatile phrases that can be easily used in such situations.
- No ImageTypes of AI learning models and main natural language AI use models
The following is a summary of the different methods of learning models used in AI and the types and characteristics of models used in major natural language-based AI.
- No ImageSummary of programming coding support AI services
A summary of major programming coding support AI services. Also includes information on the differences between services and points to consider when selecting a service.
- No ImageScenes in which interactive AI should and should not be used
Although interactive AI can be used in a variety of situations, there are some situations in which it should be used and some situations in which it should not be used. In this issue, we summarize each of these scenarios, specific examples, and points to note.
- No Image[ChatGPT/Bing chat] Risks of using interactive AI you should know about
Interactive AI is easy and free for anyone to use, but before using it for business or other purposes, you need to know that there are risks.
- No ImageBefore using ChatGPT, you need to know about its pricing plan, billing system, and usage restrictions.
ChatGPT is free to use, but there are also billing plans, so you can use ChatGPT more comfortably by knowing the difference between ChatGPT's pricing plans, billing system, and usage restrictions.
- No ImageDifference between ChatGPT and Bing Chat and the best way to use them
We have summarized the differences between ChatGPT and Bing Chat, both of which are interactive AIs that are free to use.
- No ImageWhat is Prompt Engineering? Advantages and Disadvantages
What is "prompt engineering" in interactive AI? Its advantages and disadvantages are explained in detail. Why is prompt engineering attracting attention now? The future potential of learning prompt engineering will also be explained.