Module diffusion_embeddings 

Diffusion Model-Based Knowledge Graph Embeddings

This module implements cutting-edge diffusion models for generating high-quality knowledge graph embeddings. Based on denoising diffusion probabilistic models (DDPMs) and score-based generative models for embedding generation.

Key innovations:

• Controllable embedding generation through conditioning
• High-quality embedding synthesis with noise scheduling
• Knowledge graph structure-aware diffusion processes
• Multi-scale embedding generation with hierarchical diffusion

Structs

AttentionBlock

Attention block

DiffusionConfig

Configuration for diffusion-based embeddings

DiffusionEmbeddingModel

Main diffusion embedding model

DiffusionUNet

U-Net model for diffusion denoising

NoiseScheduler

Noise scheduler for diffusion process

ResNetBlock

ResNet block for U-Net

TimeEmbedding

Time embedding for diffusion timesteps

Enums

BetaSchedule

Beta schedule types for noise scheduling

ConditioningType

Conditioning types for controlled generation

PredictionType

Types of noise prediction