Summary of Badclm: Backdoor Attack in Clinical Language Models For Electronic Health Records, by Weimin Lyu et al.
BadCLM: Backdoor Attack in Clinical Language Models for Electronic Health Records
by Weimin Lyu, Zexin Bi, Fusheng Wang, Chao Chen
First submitted to arxiv on: 6 Jul 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 explores the potential vulnerabilities of clinical language models used in electronic health records (EHR) for clinical decision support. The authors introduce an innovative attention-based backdoor attack method, BadCLM (Bad Clinical Language Models), which can embed a backdoor within the models to produce incorrect predictions when a specific trigger is present in inputs. The authors demonstrate the effectiveness of BadCLM on an in-hospital mortality prediction task using the MIMIC III dataset. This highlights a significant security risk in clinical decision support systems, emphasizing the need for future research to fortify clinical language models against such vulnerabilities. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how clinical language models used in electronic health records (EHR) can be hacked. The authors create a special kind of attack that can trick these models into making bad predictions when certain information is present. They show this works on a task predicting hospital mortality and say it’s a big problem for people using these models to make decisions. |
Keywords
» Artificial intelligence » Attention
