Summary of Task Vectors Are Cross-modal, by Grace Luo et al.
Task Vectors are Cross-Modal
by Grace Luo, Trevor Darrell, Amir Bar
First submitted to arxiv on: 29 Oct 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Computation and Language (cs.CL); 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 We investigate the internal representations of vision-and-language models (VLMs) and how they encode task representations. Our findings suggest that conceptually similar tasks are mapped to similar task vector representations, regardless of how they are specified. We identify three distinct phases in which tokens in VLMs undergo: input, task, and answer. These phases are consistent across different modalities and specifications. Additionally, we find that ensembling exemplar and instruction-based task vectors produce better task representations. Our insights shed light on the underlying mechanisms of VLMs, particularly their ability to represent tasks in a shared manner across different modalities and task specifications. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how computers understand tasks like answering questions or recognizing images. We found that when computers are given similar tasks, they use similar ways to represent those tasks, no matter what type of information is given. Computers have three steps to complete a task: understanding the input, figuring out what’s needed, and giving an answer. This process works the same way for different types of inputs like text or images. We also found that combining different ways of specifying tasks helps computers better understand those tasks. Overall, this research helps us understand how computers think about tasks and how they can do it in a similar way across different situations. |
