1 code implementation • 30 May 2024 • Chengwei Dai, Kun Li, Wei Zhou, Songlin Hu
Further analysis shows that EDIT can generate high-quality CoTs with more correct key reasoning steps.
1 code implementation • 30 May 2024 • Chengwei Dai, Kun Li, Wei Zhou, Songlin Hu
Specifically, by restructuring the training objectives -- removing the answer from outputs and concatenating the question with the rationale as input -- CasCoD's two-step learning process ensures that students focus on learning rationales without interference from the preset answers, thus improving reasoning generalizability.
1 code implementation • 22 Oct 2023 • Minxuan Lv, Chengwei Dai, Kun Li, Wei Zhou, Songlin Hu
Neural network models are vulnerable to adversarial examples, and adversarial transferability further increases the risk of adversarial attacks.
1 code implementation • 21 Oct 2023 • Chengwei Dai, Minxuan Lv, Kun Li, Wei Zhou
We study model extraction attacks in natural language processing (NLP) where attackers aim to steal victim models by repeatedly querying the open Application Programming Interfaces (APIs).