Summary of Attnlrp: Attention-aware Layer-wise Relevance Propagation For Transformers, by Reduan Achtibat et al.
AttnLRP: Attention-Aware Layer-Wise Relevance Propagation for Transformers
by Reduan Achtibat, Sayed Mohammad Vakilzadeh Hatefi, Maximilian Dreyer, Aakriti Jain, Thomas Wiegand, Sebastian Lapuschkin, Wojciech Samek
First submitted to arxiv on: 8 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); 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 novel method for attributing transformer models’ internal reasoning process has been proposed, addressing the challenges of biased predictions and hallucinations. The approach extends Layer-wise Relevance Propagation to handle attention layers, enabling faithful attribution of input and latent representations while maintaining computational efficiency similar to a single backward pass. Evaluations on LLaMa 2, Mixtral 8x7b, Flan-T5, and vision transformer architectures demonstrate the method’s superiority over existing alternatives in terms of faithfulness, opening up concept-based explanations. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Transformer models are powerful tools for processing language, but they can be prone to making biased predictions or “seeing” things that aren’t there. To understand why this is happening, researchers needed a way to see inside the model’s thinking process. A new method has been developed to do just that, by tracing how the model processes information and what it uses for guidance. This approach helps ensure that the model’s predictions are fair and accurate. |
Keywords
* Artificial intelligence * Attention * Llama * T5 * Transformer * Vision transformer
