ICLR 2023（投稿） | 扩散模型相关论文分类整理

AI Box专栏

作者：张高玮，中国人民大学高瓴人工智能学院硕士一年级，导师为赵鑫教授。

导读

ICLR是人工智能领域顶级会议之一，会议主题包括深度学习、统计和数据科学，以及一些重要的应用，例如：计算机视觉、计算生物学、语音识别、文本理解、游戏和机器人等。ICLR 2023将于2023年5月1日至5月5日在卢旺达基加利举行。官方的论文接受列表尚未公开，从投稿论文来看，扩散模型依然热度不减，是出现频率较高，且平均评分也较高的热点之一。

本文选取了与扩散模型相关的100多篇论文，按照不同的研究主题进行了分类整理，以供参考。ICLR 2023投稿论文openreview链接如下：

1. 高效采样

Dynamic Scheduled Sampling with Imitation Loss for Neural Text Generation
Truncated Diffusion Probabilistic Models and Diffusion-based Adversarial Auto-Encoders
Denoising Diffusion Samplers
Denoising MCMC for Accelerating Diffusion-Based Generative Models
DPM-Solver++: Fast Solver for Guided Sampling of Diffusion Probabilistic Models
Quasi-Taylor Samplers for Diffusion Generative Models based on Ideal Derivatives
Fast Sampling of Diffusion Models with Exponential Integrator
Accelerating Guided Diffusion Sampling with Splitting Numerical Methods
Boomerang: Local sampling on image manifolds using diffusion models
Markup-to-Image Diffusion Models with Scheduled Sampling

2. 和其它生成模型结合

Diffusion-GAN: Training GANs with Diffusion
in Conversation based on offline reinforcement learning
FastDiff 2: Dually Incorporating GANs into Diffusion Models for High-Quality Speech Synthesis
Reduce, Reuse, Recycle: Compositional Generation with Energy-Based Diffusion Models and MCMC
Geometric Networks Induced by Energy Constrained Diffusion
Progressive Image Synthesis from Semantics to Details with Denoising Diffusion GAN
Flow Matching for Generative Modeling
SPI-GAN: Denoising Diffusion GANs with Straight-Path Interpolations
Building Normalizing Flows with Stochastic Interpolants
Guiding Energy-based Models via Contrastive Latent Variables
Your Denoising Implicit Model is a Sub-optimal Ensemble of Denoising Predictions
Thinking fourth dimensionally: Treating Time as a Random Variable in EBMs

3. 在CV、NLP领域的应用

Novel View Synthesis with Diffusion Models
Pyramidal Denoising Diffusion Probabilistic Models
Compositional Image Generation and Manipulation with Latent Diffusion Models
Towards the Detection of Diffusion Model Deepfakes
DifFace: Blind Face Restoration with Diffused Error Contraction
Restoration based Generative Models
Generative Modelling with Inverse Heat Dissipation
Deep Watermarks for Attributing Generative Models
Learning multi-scale local conditional probability models of images
Images as Weight Matrices: Sequential Image Generation Through Synaptic Learning Rules
Self-conditioned Embedding Diffusion for Text Generation
Sequence to sequence text generation with diffusion models
DiffusER: Diffusion via Edit-based Reconstruction
SDMuse: Stochastic Differential Music Editing and Generation via Hybrid Representation
Universal Speech Enhancement with Score-based Diffusion
Score-based Generative 3D Mesh Modeling
CAN: A simple, efficient and scalable contrastive masked autoencoder framework for learning visual representations
Neural Volumetric Mesh Generator
SketchKnitter: Vectorized Sketch Generation with Diffusion Models
$DDM^2$: Self-Supervised Diffusion MRI Denoising with Generative Diffusion Models
Neural Image Compression with a Diffusion-based Decoder
Lossy Compression with Gaussian Diffusion
Distilling Model Failures as Directions in Latent Space
Lossy Image Compression with Conditional Diffusion Models
Quantized Compressed Sensing with Score-Based Generative Models
Out-of-distribution Detection with Diffusion-based Neighborhood

4. 在多模态领域的应用

DreamFusion: Text-to-3D using 2D Diffusion
Diffusion-based Image Translation using disentangled style and content representation
CUSTOMIZING PRE-TRAINED DIFFUSION MODELS FOR YOUR OWN DATA
Human Motion Diffusion Model
Prosody-TTS: Self-Supervised Prosody Pretraining with Latent Diffusion For Text-to-Speech
Text-Guided Diffusion Image Style Transfer with Contrastive Loss Fine-tuning
Meta-Learning via Classifier(-free) Guidance
KNN-Diffusion: Image Generation via Large-Scale Retrieval
DiffEdit: Diffusion-based semantic image editing with mask guidance
Training-Free Structured Diffusion Guidance for Compositional Text-to-Image Synthesis
Discrete Contrastive Diffusion for Cross-Modal Music and Image Generation
Unified Discrete Diffusion for Simultaneous Vision-Language Generation
ResGrad: Residual Denoising Diffusion Probabilistic Models for Text to Speech
Re-Imagen: Retrieval-Augmented Text-to-Image Generator
Prompt-to-Prompt Image Editing with Cross-Attention Control

5. 与强化学习结合

Diffusion Policies as an Expressive Policy Class for Offline Reinforcement Learning
Offline Reinforcement Learning via High-Fidelity Generative Behavior Modeling
Provably Efficient Reinforcement Learning for Online Adaptive Influence Maximization
Variational Reparametrized Policy Learning with Differentiable Physics
Is Conditional Generative Modeling all you need for Decision Making?

6. 分子图建模

Diffusion Probabilistic Modeling of Protein Backbones in 3D for the motif-scaffolding problem
Protein structure generation via folding diffusion
Pre-training Protein Structure Encoder via Siamese Diffusion Trajectory Prediction
DiffDock: Diffusion Steps, Twists, and Turns for Molecular Docking
Pocket-specific 3D Molecule Generation by Fragment-based Autoregressive Diffusion Models
Equivariant 3D-Conditional Diffusion Models for Molecular Linker Design
Structure-based Drug Design with Equivariant Diffusion Models
Equivariant Energy-Guided SDE for Inverse Molecular Design
3D Equivariant Diffusion for Target-Aware Molecule Generation and Affinity Prediction
Exploring Chemical Space with Score-based Out-of-distribution Generation
Protein Sequence and Structure Co-Design with Equivariant Translation

7. 扩散模型理论与理解

Information-Theoretic Diffusion
Analyzing diffusion as serial reproduction
Sampling is as easy as learning the score: theory for diffusion models with minimal data assumptions
Diffusion Models Already Have A Semantic Latent Space
Unifying Diffusion Models' Latent Space, with Applications to CycleDiffusion and Guidance
Understanding DDPM Latent Codes Through Optimal Transport
Interpreting Neural Networks Through the Lens of Heat Flow
gDDIM: Generalized denoising diffusion implicit models

8.扩散模型泛化与拓展

Soft Diffusion: Score Matching For General Corruptions
Where to Diffuse, How to Diffuse and How to get back: Learning in Multivariate Diffusions
Blurring Diffusion Models
Diffusion Probabilistic Fields
Neural Diffusion Processes
Pseudoinverse-Guided Diffusion Models for Inverse Problems
Removing Structured Noise with Diffusion Models
f-DM: A Multi-stage Diffusion Model via Progressive Signal Transformation
Iterative α-(de)Blending: Learning a Deterministic Mapping Between Arbitrary Densities
Score-Based Graph Generative Modeling with Self-Guided Latent Diffusion
Self-Guided Diffusion Models
From Points to Functions: Infinite-dimensional Representations in Diffusion Models
Score Matching via Differentiable Physics
Approximated Anomalous Diffusion: Gaussian Mixture Score-based Generative Models
Action Matching: A Variational Method for Learning Stochastic Dynamics from Samples
Autoregressive Generative Modeling with Noise Conditional Maximum Likelihood Estimation
DIFFUSION GENERATIVE MODELS ON SO(3)
Diffusion Posterior Sampling for General Noisy Inverse Problems

9.扩散模型迁移

Transferring Pretrained Diffusion Probabilistic Models
Cold Diffusion: Inverting Arbitrary Image Transforms Without Noise
Dual-Domain Diffusion Based Progressive Style Rendering towards Semantic Structure Preservation
Dual Diffusion Implicit Bridges for Image-to-Image Translation
Learning to Learn with Generative Models of Neural Network Checkpoints
Zero-Shot Image Restoration Using Denoising Diffusion Null-Space Model

10.特殊结构数据的建模

Autoregressive Diffusion Model for Graph Generation
Analog Bits: Generating Discrete Data using Diffusion Models with Self-Conditioning
TabDDPM: Modelling Tabular Data with Diffusion Models
ChiroDiff: Modelling chirographic data with Diffusion Models
Modeling Temporal Data as Continuous Functions with Process Diffusion
Domain Specific Denoising Diffusion Probabilistic Models for Brain Dynamics
Discrete Predictor-Corrector Diffusion Models for Image Synthesis
Diffusion-based point cloud generation with smoothness constraints
Computational Doob h-transforms for Online Filtering of Discretely Observed Diffusions
Imitating Human Behaviour with Diffusion Models
Learning Diffusion Bridges on Constrained Domains
DiGress: Discrete Denoising diffusion for graph generation
Score-based Continuous-time Discrete Diffusion Models
Brain Signal Generation and Data Augmentation with a Single-Step Diffusion Probabilistic Model

11. 鲁棒性与稳定性

DensePure: Understanding Diffusion Models towards Adversarial Robustness
Defending against Adversarial Audio via Diffusion Model
PointDP: Diffusion-driven Purification against 3D Adversarial Point Clouds
Diffusion Adversarial Representation Learning for Self-supervised Vessel Segmentation
Improving Adversarial Robustness by Contrastive Guided Diffusion Process
Robustness for Free: Adversarially Robust Anomaly Detection Through Diffusion Model
(Certified!!) Adversarial Robustness for Free!
The Biased Artist: Exploiting Cultural Biases via Homoglyphs in Text-Guided Image Generation Models
Expected Perturbation Scores for Adversarial Detection
Input Perturbation Reduces Exposure Bias in Diffusion Models
Stable Target Field for Reduced Variance Score Estimation

12.扩散模型的隐私保护

Membership Inference Attacks Against Text-to-image Generation Models
Differentially Private Diffusion Models

13. 其它方向

OCD: Learning to Overfit with Conditional Diffusion Models
Denoising Diffusion Error Correction Codes
Neural Lagrangian Schrodinger Bridge: Diffusion Modeling for Population Dynamics
Diffusion Models for Causal Discovery via Topological Ordering
Transport with Support: Data-Conditional Diffusion Bridges
A Score-Based Model for Learning Neural Wavefunctions

编辑于 2022-11-15 14:19 ・IP 属地北京

人工智能

ICLR

扩散模型