Summary of Deep Learning Based Dense Retrieval: a Comparative Study, by Ming Zhong et al.
Deep Learning Based Dense Retrieval: A Comparative Study
by Ming Zhong, Zhizhi Wu, Nanako Honda
First submitted to arxiv on: 27 Oct 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 A novel investigation assesses the vulnerability of dense retrievers to tokenizer poisoning, examining the performance degradation of BERT, DPR, Contriever, SimCSE, and ANCE models. The study finds that supervised models like BERT and DPR are significantly impacted by compromised tokenizers, while unsupervised models like ANCE exhibit greater resilience. Experiment results indicate that even slight perturbations can severely affect retrieval accuracy, emphasizing the need for robust defenses in mission-critical applications. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Dense retrievers have gotten very good at finding information, but someone might try to mess with them. This study looks into how well these systems do when their “tokenizer” (the thing that breaks down words) is tricked. They tested famous models like BERT and a few others. The results show that some models get really bad if the tokenizer is messed with, while others can handle it pretty well. It’s kind of like how you might not be able to recognize a picture if someone draws a mustache on your friend – even small changes can make a big difference. |
Keywords
» Artificial intelligence » Bert » Supervised » Tokenizer » Unsupervised
