There are always various terms in the AI field that are difficult to distinguish and understand. Today, I will make a summary, which will also help me with future lookups.
: Describe the different types of artificial intelligence and their stages of development.
Artificial Intelligence (AI): "Intelligence" demonstrated by systems created by humans, including learning, understanding, reasoning, perception, language recognition, etc. Artificial General Intelligence (AGI): Refers to the ability of a machine to possess all levels of human intelligence. Artificial Super Intelligence (ASI): Machine intelligence that surpasses the smartest and most creative levels of humans. Explainable AI (XAI): A subfield of AI focused on creating transparent models that provide clear and understandable explanations for their decisions. Chatbot: A program capable of communicating with humans through text or voice. Agents: Entities that act and interact within an environment, such as agents in reinforcement learning. Expert Systems: Computer systems that simulate the knowledge and judgment abilities of a human expert. Emergence/Emergent Behavior: A new behavior or property that arises in a complex system, which does not exist in the individual parts of the system. Generative AI: A type of AI that can create new content, such as music, articles, images, etc.
: Involve various types of machine learning methods and techniques.
Machine Learning (ML): A method for implementing artificial intelligence where machines improve their performance by learning from and understanding data. Supervised Learning: A type of machine learning where the model is provided with labeled training data. Unsupervised Learning: A type of machine learning where the model is not provided with labeled training data and must identify patterns in the data on its own. Reinforcement Learning: A type of machine learning where an agent learns to make decisions by taking actions in an environment to maximize some reward. Deep Learning: A technique for implementing machine learning that involves learning through the simulation of neural networks in the human brain. Transfer Learning: A method in machine learning that uses a pre-trained model for a new problem. Zero-shot Learning: A type of machine learning where the model makes predictions on conditions not seen during training, without any fine-tuning.
: Specific frameworks and models for building and training machine learning models.
TensorFlow: An open-source machine learning platform developed by Google for building and training machine learning models. Large Language Model (LLM): A model trained on a large amount of text data, used for understanding and generating human language. Neural Network: A computational model that mimics the working mechanism of the human brain, used for data analysis and prediction. Attention: A mechanism in neural network models used to assign different levels of importance when processing information. Transformer: A specific type of neural network architecture mainly used for processing sequential data such as natural language. Convolutional Neural Network (CNN): A form of deep learning widely applied in image processing. It automatically and effectively learns local image features through convolution operations.
Recurrent Neural Network (RNN): A type of network structure in machine learning particularly suitable for processing and predicting sequential data. The neurons in RNNs not only receive the influence of the current input but also retain historical information from previous inputs, forming a network structure with "memory" functionality.
Generative Adversarial Network (GAN): A deep learning model and one of the most promising approaches for unsupervised learning on complex distributions in recent years. It is a training method that allows two neural networks to learn by competing with each other. GPT (Generative Pretrained Transformer): A pre-trained generative Transformer model developed by OpenAI, used for various natural language processing tasks. CLIP (Contrastive Language-Image Pre-Training): A model developed by OpenAI that can understand natural language instructions and generate corresponding images.
: Techniques and methods describing how to train and optimize machine learning models.
Training Data: The dataset used to train a machine learning model. Test Data: Data used for detecting model construction, which is only utilized during model validation to evaluate the accuracy of the model. It must not be used in the model construction process, as this would lead to overfitting. Validation Data: A subset of the dataset in machine learning used to adjust hyperparameters of the model, separate from the training and test datasets. Gradient Descent: An optimization algorithm used to find the minimum value of a function, commonly applied in the training process of machine learning and deep learning models. Back Propagation: A method for calculating gradients in neural networks, often used in the training of neural networks. Fine-tuning: A process of adjusting a pre-trained model to adapt to a new task. Alignment: The degree to which an AI's behavior is consistent with its designer's intentions. Prompt: Input information used to trigger specific output generation from an AI model. Chain of Thought (CoT): Refers to the thinking process of an AI model when processing information and making decisions.
：Describe how to evaluate the performance of a machine learning model and the possible issues that may arise.
Overfitting: When a statistical model or machine learning algorithm matches the training data too closely, resulting in decreased performance on new data. Underfitting: A modeling error that occurs when a statistical model or machine learning algorithm cannot adequately capture the underlying structure of the data. Data Augmentation: A method to reduce model overfitting by transforming and expanding the training data to enhance the model's generalization ability. Regularization: In machine learning, regularization is a technique used to prevent overfitting by adding a penalty term to the model's loss function.
: Describes the components of a machine learning model and how to adjust these components to improve model performance.
Parameters: Variables learned through data during the model training process, such as the weights in a neural network. Hyperparameter Tuning: The process of finding the optimal hyperparameters to maximize model performance. Hidden Layer: A layer between the input and output layers in a neural network, used to extract various features from the input.
: A hardware device designed to accelerate compute-intensive tasks, such as machine learning.
GPU (Graphics Processing Unit): A hardware device used to accelerate graphics and image processing tasks, also commonly used for deep learning computations. TPU (Tensor Processing Unit): A type of microprocessor specifically developed by Google to accelerate machine learning workloads. Accelerator: A microprocessor specifically designed to accelerate AI applications, which can significantly improve processing speed and efficiency. Compute(计算)：The computing resources used to perform AI tasks.
Stable Diffusion Related
Stable Diffusion: An image generation model published by CompVis in 2022, consisting of U-Net, VAE, and a text encoder, used to create specific types of images. Model (Checkpoint): Also known as checkpoint files, these are pre-trained stable diffusion weights designed to create general or specific types of images. (Low-Rank Adaptation, LoRA): LoRAs are smaller files (usually 10-200MB) that are used in conjunction with existing stable diffusion checkpoint models to introduce new themes into images, which can range from characters to art styles to poses. VAE Aesthetic Embedding: A method that improves and repairs details, saturation, and other aspects of images. Hypernetwork: Also known as a style model, this is a fine-tuning technique that attaches a small neural network to modify styles. ): A deep learning model published by Google Research and Boston University in 2022, used to fine-tune existing image generation models. Text inversion: A type of embedding training method used to extract and learn textual information. Safe tensors (safetensors): A model format developed by Huggingface for securely saving tensor data.