Summary of How Lightweight Can a Vision Transformer Be, by Jen Hong Tan
How Lightweight Can A Vision Transformer Be
by Jen Hong Tan
First submitted to arxiv on: 25 Jul 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- 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 paper proposes a novel strategy for vision transformers using Mixture-of-Experts (MoE) to streamline, rather than augment, vision transformer models. The approach utilizes SwiGLU feedforward networks as experts in each MoE layer, without complex attention or convolutional mechanisms. Depth-wise scaling is applied to reduce the size of the hidden layer and increase the number of experts. Grouped query attention is also employed. The proposed architecture is competitive even at a size of 0.67M parameters. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper shows that MoE can be used to simplify vision transformer models, making them more efficient and easier to train. By using feedforward networks as experts, the model doesn’t need complex attention mechanisms or convolutional layers. This makes it possible to use the architecture even with limited computational resources. The results show that the proposed approach is competitive with other state-of-the-art models at a much smaller size. |
Keywords
* Artificial intelligence * Attention * Mixture of experts * Vision transformer
