Summary of Zero-shot Generalization Across Architectures For Visual Classification, by Evan Gerritz et al.
Zero-shot generalization across architectures for visual classification
by Evan Gerritz, Luciano Dyballa, Steven W. Zucker
First submitted to arxiv on: 21 Feb 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 A recent study investigates the relationship between deep learning networks’ ability to generalize to unseen data and their classification accuracy. The researchers used a minimalist vision dataset and a measure of generalizability to evaluate popular network architectures, including convolutional neural networks (CNNs) and transformers. They found that different networks vary in their power to extrapolate to new classes, both across layers and architectures. Surprisingly, accuracy is not the best predictor of generalization, and the ability to generalize can even decrease with increasing layer depth. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A group of scientists studied how well deep learning networks work on new data they’ve never seen before. They used a simple image dataset and measured how good each network was at recognizing objects it had never seen before. They found that different types of networks are better or worse at this task, depending on the layer it’s in and the type of architecture. What’s surprising is that just because a network is really good at classifying images doesn’t mean it’ll do well on new ones. |
Keywords
* Artificial intelligence * Classification * Deep learning * Generalization
