Summary of Lifelong Learning Of Video Diffusion Models From a Single Video Stream, by Jason Yoo et al.
Lifelong Learning of Video Diffusion Models From a Single Video Stream
by Jason Yoo, Yingchen He, Saeid Naderiparizi, Dylan Green, Gido M. van de Ven, Geoff Pleiss, Frank Wood
First submitted to arxiv on: 7 Jun 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary In this paper, researchers show that autoregressive video diffusion models can be trained from a single continuous video stream, achieving competitive results with traditional offline training methods despite using the same number of gradient steps. The study also explores experience replay, which retains only a subset of preceding frames, and demonstrates its effectiveness in lifelong learning settings. To evaluate this approach, the authors introduce three new datasets for generative modeling: Lifelong Bouncing Balls, Lifelong 3D Maze, and Lifelong PLAICraft, each containing over a million consecutive frames from increasingly complex environments. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper trains video models using a single video stream and shows that it can be as good as training them offline. The researchers also use a technique called experience replay to make the model learn better over time. To test this approach, they created three new datasets for training video generative models: Bouncing Balls, 3D Maze, and PLAICraft. |
Keywords
* Artificial intelligence * Autoregressive
