Summary of Outdated Issue Aware Decoding For Reasoning Questions on Edited Knowledge, by Zengkui Sun et al.
Outdated Issue Aware Decoding for Reasoning Questions on Edited Knowledge
by Zengkui Sun, Yijin Liu, Jiaan Wang, Fandong Meng, Jinan Xu, Yufeng Chen, Jie Zhou
First submitted to arxiv on: 5 Jun 2024
Categories
- Main: Computation and Language (cs.CL)
- 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| GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes a novel decoding strategy, called outDated ISsue aware deCOding (DISCO), to improve the performance of edited models on reasoning questions in Knowledge Editing. The existing methods tend to memorize superficial word compositions rather than truly learning and absorbing the updated knowledge, leading to outdated responses. DISCO captures the difference in probability distributions between original and edited models, amplifies the token prediction difference, and reduces the ratio of outdated issues by 5.78% on the zsRE dataset. The proposed method outperforms the previous state-of-the-art (SOTA) method by 12.99 F1 scores. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary In this paper, researchers are trying to fix a problem with how computers learn new information. Right now, when we update what a computer knows, it just memorizes the new words instead of really understanding them. This means that when the computer tries to answer questions based on its updated knowledge, it often gives outdated answers. The team proposes a simple way to make computers give better answers by looking at how different the new information is from what they already knew. This helps them give more accurate answers and reduces the number of outdated answers. |
Keywords
» Artificial intelligence » Probability » Token
