TMLab: Generative Enhanced Model (GEM) for adversarial attacks
This work addresses the challenge of improving adversarial robustness in fact-checking systems, though it appears incremental as it builds upon existing GPT-2 architecture with targeted modifications.
The paper tackled the problem of generating adversarial claims to challenge fact-checking systems, resulting in a model that won first prize on the FEVER 2.0 Breakers Task by creating malicious claims that mixed facts from various articles, making them difficult to classify for truthfulness.
We present our Generative Enhanced Model (GEM) that we used to create samples awarded the first prize on the FEVER 2.0 Breakers Task. GEM is the extended language model developed upon GPT-2 architecture. The addition of novel target vocabulary input to the already existing context input enabled controlled text generation. The training procedure resulted in creating a model that inherited the knowledge of pretrained GPT-2, and therefore was ready to generate natural-like English sentences in the task domain with some additional control. As a result, GEM generated malicious claims that mixed facts from various articles, so it became difficult to classify their truthfulness.