Summary of On the Effects Of Data Scale on Ui Control Agents, by Wei Li et al.
On the Effects of Data Scale on UI Control Agents
by Wei Li, William Bishop, Alice Li, Chris Rawles, Folawiyo Campbell-Ajala, Divya Tyamagundlu, Oriana Riva
First submitted to arxiv on: 6 Jun 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- 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 This paper investigates the effectiveness of fine-tuning Large Language Models (LLMs) to power autonomous agents controlling computer interfaces. The authors collect a new dataset, AndroidControl, comprising 15,283 demonstrations of everyday tasks with Android apps. Each task instance includes high and low-level human-generated instructions, allowing exploration of task complexity. The dataset is the most diverse to date, featuring 14,548 unique tasks across 833 Android apps. The study finds that fine-tuned models outperform zero-shot and few-shot baselines when tested in-domain, but scaling out-of-domain suggests that collecting more data alone may be insufficient for achieving robust performance, particularly for high-level tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research looks at how well computers can learn to control apps on our phones. The scientists created a big dataset with lots of examples of people doing everyday tasks on their Android devices. They wanted to see if just giving the computer more data would make it better at controlling the phone. They found that when tested with the same kind of data they learned from, the computer did get better. But when they tested it with new, different data, it didn’t improve as much. This means that we might need to come up with other ways to help computers learn these skills. |
Keywords
» Artificial intelligence » Few shot » Fine tuning » Zero shot
