Summary of Peter Parker or Spiderman? Disambiguating Multiple Class Labels, by Nuthan Mummani et al.
Peter Parker or Spiderman? Disambiguating Multiple Class Labels
by Nuthan Mummani, Simran Ketha, Venkatakrishnan Ramaswamy
First submitted to arxiv on: 25 Oct 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
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 proposes a framework to disambiguate two distinct possibilities in deep network predictions during inference, where two top-k predictions are made. The first possibility is that each prediction is driven by different sets of entities in the input, while the second possibility is that a single entity drives both predictions, effectively making two separate guesses about the same entity type. Current interpretability techniques do not address this issue, as they focus on one class label at a time. The authors introduce a method combining modern segmentation and input attribution techniques to resolve these ambiguities, providing a simple counterfactual “proof” for each case. This framework is demonstrated to be effective on the ImageNet validation set with multiple models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us better understand how deep networks make predictions. When a network makes two predictions, there are two possibilities: either they’re driven by different things in the input or by the same thing. Right now, we don’t have good ways to figure out which is true. The authors of this paper came up with a new way to do just that, using techniques from computer vision and attribution methods. This helps us understand what’s going on inside these powerful networks. |
Keywords
» Artificial intelligence » Inference
