Summary of Cosy: Evaluating Textual Explanations Of Neurons, by Laura Kopf et al.
CoSy: Evaluating Textual Explanations of Neurons
by Laura Kopf, Philine Lou Bommer, Anna Hedström, Sebastian Lapuschkin, Marina M.-C. Höhne, Kirill Bykov
First submitted to arxiv on: 30 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
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 presents a novel approach called CoSy (Concept Synthesis) to evaluate textual explanations of latent neurons in Deep Neural Networks (DNNs). The authors aim to address the lack of a unified quantitative method to assess the quality of these explanations. They propose a generative model conditioned on textual input that generates data points representing the explanations, and compare them with control data points to estimate the explanation’s quality. The framework is evaluated through sanity checks and benchmarked against various neuron description methods for Computer Vision tasks, revealing significant differences in quality. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us understand how deep neural networks (DNNs) learn from pictures and words. It’s hard to tell if a DNN really understands what it’s learning because we can’t easily explain why it made certain decisions. The researchers created a new way called CoSy (Concept Synthesis) to measure how well a DNN explains its thoughts in words. They used a special model that generates text based on the DNN’s ideas and compared those with random texts. This showed which explanations were better than others. The team tested their method on different ways of explaining DNNs’ thought processes for image recognition tasks, and found some methods were much better than others. |
Keywords
» Artificial intelligence » Generative model
