Summary of Locally Testing Model Detections For Semantic Global Concepts, by Franz Motzkus et al.
Locally Testing Model Detections for Semantic Global Concepts
by Franz Motzkus, Georgii Mikriukov, Christian Hellert, Ute Schmid
First submitted to arxiv on: 27 May 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); 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 proposed framework, called global-to-local Concept Attribution (glCA), aims to ensure the quality of black-box Deep Neural Networks (DNNs) by linking global concept encodings to local processing. glCA combines approaches from local and global Explainable Artificial Intelligence (xAI) to test DNNs for predefined semantical concepts. This approach enables conditioning post-hoc explanations on semantic concepts encoded as linear directions in the model’s latent space, allowing for pixel-exact scoring of global concept usage. The results show significant differences in local perception and usage of individual global concept encodings, highlighting the importance of thorough semantic concept encodings. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary In a nutshell, researchers have created a new way to understand how deep learning models work. They want to make sure these models are reliable and safe, especially for important tasks like self-driving cars. The new method looks at how individual parts of the model process information related to specific concepts, like “car” or “person.” This helps experts see what’s going on inside the model and why it makes certain predictions. |
Keywords
» Artificial intelligence » Deep learning » Latent space
