Summary of Backward Lens: Projecting Language Model Gradients Into the Vocabulary Space, by Shahar Katz et al.
Backward Lens: Projecting Language Model Gradients into the Vocabulary Space
by Shahar Katz, Yonatan Belinkov, Mor Geva, Lior Wolf
First submitted to arxiv on: 20 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- 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 paper tackles the challenge of understanding how Transformer-based Language Models (LMs) learn and recall information. By extending existing interpretability methods, researchers can now analyze not only the forward pass but also the backward pass and gradients within these models. The study demonstrates that gradient matrices can be decomposed into a low-rank linear combination of their inputs from both passes, allowing for novel insights into how new information is stored in LMs’ neurons. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Language Models are powerful tools that help us understand language. But have you ever wondered how they learn and remember things? A team of researchers wanted to figure this out, so they developed a new way to look at how these models work. They took existing methods and adapted them to also analyze the backward pass and gradients within the model. This allowed them to see how new information is stored in the neurons of the Language Model. |
Keywords
* Artificial intelligence * Language model * Recall * Transformer
