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

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

B

BERT Backpropagation Batch Normalization Boosting Bias

C

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

D

Deepfake Deep Learning Data Augmentation Deterministic Model Discriminative Model

E

Epoch Ensemble Learning Encoder Explainable AI (XAI)Embedding

F

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

G

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

H

Hyperparameter Heuristic Hidden Layer Hierarchical Model Hallucination

I

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

J

JAX JSONL / JSON-lines Jittering Joint Embedding Juxtaposition

K

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

L

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

M

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

N

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

O

One-hot Encoding Overfitting Optimizer Online Learning Objective Function

P

Pooling Parameter Pretraining Prompt Policy / Reinforcement Learning Policy

Q

Q-learning Queue / Buffer Query Quality Estimation Quantization

R

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

S

Softmax Sampling Sequence Modeling Supervised Learning Self-Supervised Learning

T

Tuning / Hyperparameter Tuning Tokenizer Transformer Training Data Transfer Learning

U

U-Net Underfitting Universal Approximation Theorem Unsupervised Learning Uncertainty Estimation

V

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

W

Weight Decay Whitening / Whitening Transformation Workflow Word Embedding Weak Supervision

X

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

Y

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

Z

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

Weight Decayとは何か

Weight Decayは、機械学習や深層学習で広く使用される正則化技術であり、過学習を防ぐことを目的としています。

これは、損失関数にペナルティ項を追加することで、大きな重みの値を抑制し、モデルがより小さな重みを学習するように促します。

この技術は、複雑なモデルや高次元データセットに特に有益であり、モデルが見たことのないデータに対してよりよく一般化できるようにします。

Weight Decayは、通常、Dropoutなどの他の正則化手法と組み合わせて使用され、モデルの堅牢性を高めます。

深層学習技術が進歩するにつれて、Weight Decayは適応型学習率やより洗練された最適化手法によってさらに進化する可能性があります。