Zero-Shot Learning (ZSL) is a machine learning scenario in which an AI model is trained to recognize and classify objects or concepts without prior knowledge of any examples of those categories or concepts. Most deep learning...
Sentiment analysis, also known as sentiment classification or opinion mining, is a natural language processing technique used to analyze text data to identify and extract the author's sentiment, opinion, and evaluation. It involves evaluating text...
A token is a credential used to identify a user. It is generated by the server and returned to the client. The client includes this token in subsequent requests, and the server verifies the user's identity by parsing the token to determine whether to authorize access to resources.
Benchmarking is a method for evaluating and comparing system performance. It measures a system's performance through a series of standardized test procedures. This type of testing helps determine a system's performance under specific conditions...
Artificial Narrow Intelligence (ANI), also known as "weak AI," refers to AI systems designed and trained to perform specific tasks or tasks within a narrow scope. It differs from systems with human-like...
Forward propagation is a core process in neural networks, describing how input data is passed through network layers and generates output. Input data is fed into the input layer of the neural network. Input data...
Deliberative alignment is a novel training method proposed by OpenAI, designed to improve the safety and reliability of large language models. This method combines process-based and outcome-based supervision to allow the model to...
Instruction tuning (IT) is a training method for large language models (LLMs) that aims to improve the model's ability to follow natural language instructions and complete real-world tasks.
Visual-language models (VLMs) are a type of multimodal artificial intelligence system that combines image and text processing capabilities to perform high-level visual language tasks, such as visual question answering.
Underfitting refers to a machine learning model performing poorly on training data, resulting in poor performance on test data. It occurs when the model is too simple and fails to capture complex patterns in the data. Underfitting...
Robustness refers to the ability of a system, model, or entity to maintain its intended function, stable performance, and accurate results when faced with disturbances, noise, outliers, or changes in design parameters in the input data.
Face recognition is a technology that identifies individuals based on their facial features. It uses computer vision and pattern recognition techniques to analyze input facial images or video streams, first determining if...
AI alignment refers to the field of ensuring that the behavior of artificial intelligence systems aligns with human intentions and values. The core objectives can be summarized by four key principles: robustness, reliability, and...
Image generation is an important research area in computer vision. It uses deep learning models, such as VAEs, GANs, and Diffusion Models, to learn from data and generate new images. This...
The Turing Test is an experiment proposed in 1950 by British mathematician and logician Alan Turing to determine whether machines can exhibit intelligence equivalent to or indistinguishable from that of humans...
Hidden layers are intermediate layers in artificial neural networks, located between the input and output layers. Their function is to extract and transform features from the input data, providing high-level features for the final output layer. Hidden layers...
The Transformer model is a deep learning model first proposed in 2017 by a Google research team in their paper "Attention is All You Need". The core idea of this model is the self-attention mechanism...
Model quantization is a key technique in deep learning model optimization. It reduces the storage and computational requirements of the model by decreasing the bit width of its parameters, thereby improving the model's performance on various hardware platforms...
NeRF (Neural Radiance Fields) is a computer vision technique for generating high-quality 3D reconstruction models. It uses deep learning techniques to extract the geometric shape and texture information of objects from images viewed from multiple perspectives, and then...
Facial recognition is a biometric technology that identifies individuals based on their facial features. It automatically detects and tracks faces by acquiring images or video streams containing human faces, and further...
3D Gaussian Splatting is an emerging 3D scene reconstruction and rendering technique. It uses 3D Gaussian functions to represent points in a scene and projects these Gaussian functions onto a 2D image plane...
Contextual embedding is a technique that maps words to vector spaces, generating a representation for each word based on its context. These representations capture the diversity of words in different contexts...
Regularization is a technique used in machine learning to prevent overfitting. It's achieved by adding a penalty term to the model's loss function, which is related to the complexity of the model parameters. Regularization helps limit...
Data augmentation is a common technique in machine learning and deep learning, widely used in computer vision and natural language processing. Its core purpose is to transform or generate new data...