Summary of Cat: Interpretable Concept-based Taylor Additive Models, by Viet Duong et al.
CAT: Interpretable Concept-based Taylor Additive Models
by Viet Duong, Qiong Wu, Zhengyi Zhou, Hongjue Zhao, Chenxiang Luo, Eric Zavesky, Huaxiu Yao, Huajie Shao
First submitted to arxiv on: 25 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 GAMs adopt neural networks to learn non-linear functions for each feature, which are then combined through a linear model for final predictions. While GAMs can explain DNNs at the feature level, they require large numbers of model parameters and are prone to overfitting, making them hard to train and scale. To address these issues, researchers have shifted towards concept-based interpretable methods that integrate concept learning as an intermediate step before making predictions. However, these methods require domain experts to extensively label concepts with relevant names and their ground-truth values. To simplify this process, we propose CAT, a novel interpretable Concept-bAsed Taylor additive model that only requires users to categorize input features into broad groups through a quick metadata review. CAT first embeds each group of input features into one-dimensional high-level concept representation, then feeds the concept representations into a new white-box Taylor Neural Network (TaylorNet) that learns the non-linear relationship between inputs and outputs using polynomials. Evaluation results demonstrate that CAT can outperform or compete with baselines while reducing model parameters. Importantly, it explains model predictions through high-level concepts that humans can understand. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary GAMs try to explain deep neural networks (DNNs) by learning non-linear functions for each feature. But this approach has some problems. It uses too many model parameters and can get stuck in a “rut” so it’s hard to train and scale. To make things better, researchers are looking at ways to explain DNNs using concepts that humans can understand. This means getting experts to label these concepts with names and values, which takes a lot of time and effort. Our solution is called CAT (Concept-bAsed Taylor). It doesn’t need all this extra work because it just asks users to group input features into simple categories. Then, it uses a special kind of neural network that can learn patterns in the data using polynomials. |
Keywords
» Artificial intelligence » Neural network » Overfitting
