Summary of Coordinate in and Value Out: Training Flow Transformers in Ambient Space, by Yuyang Wang et al.
Coordinate In and Value Out: Training Flow Transformers in Ambient Space
by Yuyang Wang, Anurag Ranjan, Josh Susskind, Miguel Angel Bautista
First submitted to arxiv on: 5 Dec 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 The abstract discusses the limitations of traditional flow matching models for generative modeling on various data modalities, such as images or videos, and unstructured 3D point clouds. These models typically require a two-stage training process involving a variational auto-encoder (VAE) followed by a flow matching generative model in the low-dimensional latent space. The authors introduce Ambient Space Flow Transformers (ASFT), a domain-agnostic approach that eliminates the need for specific data compressors and simplifies the training process. ASFT employs general-purpose transformer blocks to make predictions continuously in coordinate space, outperforming comparable approaches on images and 3D point clouds. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Flow matching models are used for generative modeling on different types of data. These models usually require a two-step process: first, a variational auto-encoder (VAE) is trained, then a flow matching generative model is trained in the VAE’s low-dimensional space. However, this makes it hard to use these models with different types of data. To fix this, researchers created Ambient Space Flow Transformers (ASFT), which can handle many types of data without needing special compressors. ASFT uses transformer blocks to make predictions and does well on both images and 3D point clouds. |
Keywords
» Artificial intelligence » Encoder » Generative model » Latent space » Transformer
