AI 용어집
인공지능 완전 사전
Language Model
Statistical or neural system that calculates the probability of word sequences appearing in a language. These models learn contextual and syntactic dependencies from large text corpora to generate or evaluate natural language.
Transformer Architecture
Neural architecture based on attention mechanisms that processes sequences in parallel without temporal dependencies. Transformers have revolutionized language models thanks to their ability to capture long-distance dependencies.
GPT
Family of generative language models based on a Transformer architecture decoding only from left to right. GPT models specialize in generating coherent text and sequence completion.
N-grams
Statistical language models based on the Markovian hypothesis that the probability of a word depends only on the previous n-1 words. N-grams constitute the classic approach to language modeling before the neural network era.
RNN
Recurrent neural network that processes sequences by maintaining a hidden state that evolves at each token. RNNs were among the first neural architectures applied to language models to capture temporal dependencies.
LSTM
Advanced variant of RNN using forget gates and memory gates to manage long-term dependencies. LSTMs overcame the vanishing gradient problem of traditional RNNs in language modeling applications.
Causal Language Models
Models trained to predict the next word based only on previous words in the sequence. Causal models are particularly suited for text generation and completion tasks.
Masked Language Models
Models trained to predict masked words in a sequence using the full bidirectional context. This approach allows for better context understanding for analysis and classification tasks.