Summary of Do Causal Predictors Generalize Better to New Domains?, by Vivian Y. Nastl and Moritz Hardt
Do causal predictors generalize better to new domains?
by Vivian Y. Nastl, Moritz Hardt
First submitted to arxiv on: 15 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Machine Learning (stat.ML)
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 Machine learning models trained on causal features do not generalize as well across domains as previously thought. Researchers tested 16 prediction tasks on tabular datasets covering various applications, such as health and education. Each dataset had multiple domains, allowing for the evaluation of how well a model performed in one domain when trained in another. The results showed that models using all available features, regardless of causality, outperformed those using causal features both in-domain and out-of-domain. Furthermore, recent causal machine learning methods did not perform better than standard predictors trained on causal features. This study highlights the limitations of using causal features for domain generalization. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Machine learning models are super smart at predicting things! But what happens when they’re asked to predict something new that’s a bit different from what they’ve seen before? Researchers looked at how well these models do when they’re trained on special features that have an effect on what they’re trying to predict. They tested the models on 16 different tasks, using data from things like health and education. What they found was that the models that use all the available features, no matter if they’re important or not, actually do better than those that only use the special features that have an effect. This is important because it helps us understand what these models are good at and what we should use them for. |
Keywords
* Artificial intelligence * Domain generalization * Machine learning
