Understanding and Mitigating Copying in Diffusion Models @GoogleTechTalks
Understanding and Mitigating Copying in Diffusion Models  @GoogleTechTalks
Uploaded December 2023 | Updated November 2026, 6 days ago
A Google TechTalk, presented by Gowthami Somepalli , 2023-08-09
Abstract: Cutting-edge diffusion models produce high-quality and customizable images, enabling their use for commercial art and graphic design purposes. First, I will discuss our study of various frameworks to detect replication. Then, I will show how we used these frameworks to identify memorization in Stable Diffusion 1.4 model. In the second part, I will discuss various factors contributing to memorization in diffusion models. While it is widely believed that duplicated images in the training set are responsible for content replication at inference time, I will show results on how the text conditioning of the model also plays an important role. Lastly, I will discuss several techniques we proposed based on these findings for reducing data replication in both training and inference times.
Understanding and Mitigating Copying in Diffusion ModelsPOPri: Private Federated Learning using Preference-Optimized Synthetic Data2023 Blockly Developer Summit Day 2-8: Active STEM with Unruly SplatsGo Meetup April 2025 - Git Bisect and Go Size AnalyzerZürich Go Meetup: Run Go programs on your Raspberry Pi with gokrazy!Leveraging Per-Instance Privacy for Machine Unlearning2023 Blockly Developer Summit Day 2-2: Blockly Themes for AccessibilityGoing Back and Beyond: Emerging (Old) Threats in LLM Privacy and PoisoningWhat Makes Software Work?Introduction to Length-Constrained Expanders and Expander DecompositionsOne Tree to Rule Them All: Polylogarithmic Universal Steiner TreesUnderstanding Trade-offs Between Worst-case Differential Privacy Guarantees & Real Threat Models
Google TechTalks |

Understanding and Mitigating Copying in Diffusion Models

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER