Diffusion Loss

Diffusion loss is a crucial component in training generative models, particularly diffusion models, aiming to accurately model the probability distribution of data points in a continuous space. Current research focuses on improving the robustness and efficiency of diffusion loss, exploring its application within various architectures like autoregressive models and joint-embedding predictive architectures (JEPAs), and addressing challenges such as the "corruption stage" observed during few-shot fine-tuning. These advancements enhance the quality and diversity of generated data across various modalities (images, audio, video) and improve the efficiency of training, leading to significant improvements in generative modeling capabilities and downstream applications.

Papers

October 24, 2024

Diffusion Attribution Score: Evaluating Training Data Influence in Diffusion Model
Jinxu Lin, Linwei Tao, Minjing Dong, Chang Xu
Diffusion Model Attribution Score Data Attribution Training Data Influence Diffusion Loss

October 2, 2024

Denoising with a Joint-Embedding Predictive Architecture
Dengsheng Chen, Jie Hu, Xiaoming Wei, Enhua Wu
Generative Modeling Self Supervised Representation Learning Joint Embedding Predictive Architecture Diffusion Loss

June 17, 2024

Autoregressive Image Generation without Vector Quantization
Tianhong Li, Yonglong Tian, He Li, Mingyang Deng, Kaiming He
Autoregressive Model Vector Quantization Autoregressive Image Generation Diffusion Loss Vector Quantized

May 30, 2024

Exploring Diffusion Models' Corruption Stage in Few-Shot Fine-tuning and Mitigating with Bayesian Neural Networks
Xiaoyu Wu, Jiaru Zhang, Yang Hua, Bohan Lyu, Hao Wang, Tao Song, Haibing Guan
Diffusion Model Fine Tuning Variational Inference Bayesian Neural Network Faulty Negative Mitigation Corruption Level Diffusion Loss Shot Fine Tuning Shot Tuning

March 25, 2024

Improving Diffusion Models's Data-Corruption Resistance using Scheduled Pseudo-Huber Loss
Artem Khrapov, Vadim Popov, Tasnima Sadekova, Assel Yermekova, Mikhail Kudinov
Diffusion Model Loss Function Reverse Diffusion Huber Loss Diffusion Loss

March 20, 2024

Editing Massive Concepts in Text-to-Image Diffusion Models
Tianwei Xiong, Yue Wu, Enze Xie, Yue Wu, Zhenguo Li, Xihui Liu
Text to Image Diffusion Model T2I Diffusion Model Categorical Structure Diffusion Loss

November 30, 2023

Brainformer: Mimic Human Visual Brain Functions to Machine Vision Models via fMRI
Xuan-Bac Nguyen, Xin Li, Pawan Sinha, Samee U. Khan, Khoa Luu
High Dimensional Functional Magnetic Resonance Imaging Computer Vision Model Diffusion Loss

October 30, 2023

DiffEnc: Variational Diffusion with a Learned Encoder
Beatrix M. G. Nielsen, Anders Christensen, Andrea Dittadi, Ole Winther
Diffusion Model Hierarchical Variational Deep Encoder Diffusion Loss Depth Function

March 21, 2023

Vox-E: Text-guided Voxel Editing of 3D Objects
Etai Sella, Gal Fiebelman, Peter Hedman, Hadar Averbuch-Elor
Latent Diffusion Model 3D Object Text to 3D Text Guided Diffusion Model Voxel Based Semantic Segmentation Text to 3D Synthesis Diffusion Loss

December 26, 2022

Unsupervised Representation Learning from Pre-trained Diffusion Probabilistic Models
Zijian Zhang, Zhou Zhao, Zhijie Lin
Representation Learning Pre Trained Diffusion Model Diffusion Probabilistic Model Unsupervised Representation Learning Auto Encoder Model Diffusion Autoencoder Diffusion Loss

July 7, 2022

An Embedding-Dynamic Approach to Self-supervised Learning
Suhong Moon, Domas Buracas, Seunghyun Park, Jinkyu Kim, John Canny
Self Supervised Learning Self Supervised Image Classification Dynamic Embeddings Diffusion Loss