Summary of Functional Faithfulness in the Wild: Circuit Discovery with Differentiable Computation Graph Pruning, by Lei Yu et al.
Functional Faithfulness in the Wild: Circuit Discovery with Differentiable Computation Graph Pruning
by Lei Yu, Jingcheng Niu, Zining Zhu, Gerald Penn
First submitted to arxiv on: 4 Jul 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: 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 The paper proposes DiscoGP, an algorithm for discovering circuits in language models (LMs). Circuit discovery is the task of breaking down LMs into sparse subnetworks, which can help understand their computational mechanisms. The authors identify two limitations in current approaches: a choice between pruning connections or weights, and algorithms that often miss essential components of complete circuits. DiscoGP addresses these issues by using differentiable masking to discover circuits. The algorithm is evaluated on benchmarks and outperforms existing methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper introduces an algorithm called DiscoGP that helps understand how language models work. It’s like taking apart a toy to see what makes it tick. Right now, there are two problems with the way we do this: we have to choose between cutting connections or weights, and sometimes we miss important parts of the model. DiscoGP is a new way to discover circuits in language models that doesn’t have these limitations. |
Keywords
* Artificial intelligence * Pruning
