Glossary

0-9

1-shot learning 2-stage detector 3D convolution 4D data 5G + AI 6DoF pose estimation 7D representation 8-bit quantization 9-layer network 0-shot learning

A

Algorithm Autoencoder Artificial Intelligence (AI)Attention AGI / Artificial General Intelligence

B

Batch Normalization Boosting Backpropagation Bias BERT

C

Classifier / Classification Chatbot Cross-Validation Clustering CNN / Convolutional Neural Network

D

Deterministic Model Data Augmentation Deep Learning Discriminative Model Deepfake

E

Explainable AI (XAI)Ensemble Learning Encoder Embedding Epoch

F

Foundation Model Fine-tuning Forward Propagation Feature Extraction Fusion / Multimodal Fusion

G

Gradient Descent Graph Neural Network (GNN)Generative AI GAN / Generative Adversarial Network Grounding

H

Hierarchical Model Hidden Layer Hyperparameter Hallucination Heuristic

I

Imbalanced Data Instance / Sample Intelligence Amplification / Augmentation Interpretability Instruction tuning

J

Juxtaposition Joint Embedding Jittering JAX JSONL / JSON-lines

K

Knowledge Distillation Kernel Trick K-means Clustering K-Shot Learning KL Divergence (Kullback–Leibler Divergence)

L

Large Language Model (LLM)Loss Function Latent Variable Learning Rate LSTM / Long Short-Term Memory

M

Machine Learning (ML)Multimodal / Multimodality Multi-head Attention Meta-learning Model

N

Novelty Detection / Anomaly Detection Neural Network Normalization NLP / Natural Language Processing NLU / Natural Language Understanding

O

Objective Function Overfitting Online Learning Optimizer One-hot Encoding

P

Parameter Policy / Reinforcement Learning Policy Prompt Pretraining Pooling

Q

Quantization Query Queue / Buffer Quality Estimation Q-learning

R

Reinforcement Learning (RL)Retrieval Augmented Generation (RAG)Regularization Representation Learning RNN / Recurrent Neural Network

S

Supervised Learning Sampling Sequence Modeling Self-Supervised Learning Softmax

T

Transfer Learning Tokenizer Tuning / Hyperparameter Tuning Transformer Training Data

U

Uncertainty Estimation Unsupervised Learning Underfitting Universal Approximation Theorem U-Net

V

Validation Set Vector Embedding Variational Autoencoder (VAE)Vanishing / Exploding Gradient Vision Transformer (ViT)

W

Whitening / Whitening Transformation Weak Supervision Word Embedding Workflow Weight Decay

X

X-axis / feature axis XAI / Explainable AI XLM XLNet XOR problem

Y

Yield (model yield / throughput)Yoga of AI Y-transform / YUV Y-axis / feature axis YAGNI (You Aren't Gonna Need It)

Z

Zero-gradient phenomenon Zero-centric / Zero-bias initialization Zero-shot Learning / Zero-shot inference Zygosity in augmentation Z-score Normalization

Weight Decay란 무엇인가

Weight Decay는 머신러닝과 딥러닝에서 과적합을 방지하기 위해 널리 사용되는 정규화 기술입니다.

이는 손실 함수에 패널티 항을 추가하여 큰 가중치 값을 억제하고 모델이 더 작은 가중치를 학습하도록 유도합니다.

이 기술은 복잡한 모델과 고차원 데이터 세트에 특히 유용하며, 이는 모델이 보지 못한 데이터에서 더 잘 일반화하도록 도와줍니다.

Weight Decay는 일반적으로 Dropout과 같은 다른 정규화 방법과 결합하여 모델의 강건성을 향상시킵니다.

딥러닝 기술이 발전함에 따라 Weight Decay는 적응형 학습률과 더 정교한 최적화 방법을 통해 further 발전할 수 있습니다.