Recent research underscores the potential hazards that Backdoor Attacks pose to natural language processing (NLP) models. A thorough exploration of these attack methodologies is critical for comprehending the susceptibility of such models. Under normal circumstances, a model compromised by a backdoor attack will produce standard outputs; however, the presence of a specific trigger within the input leads to erroneous results. This paper focuses on the vulnerability of BERT, a widely recognized model in numerous NLP applications, by introducing a novel backdoor attack strategy that effectively compromises it. We manipulate the attention heads in BERT to enhance the backdoor attack. The efficacy of this method is demonstrated through experiments conducted on clean-label attack and a Sentiment Analysis task.

Language models like BERT dominate current NLP research due to their robust performance, but they are vulnerable to backdoor attacks. Such attacks cause the model to consistently generate incorrect predictions when specific triggers are present in the input, while maintaining normal behavior on clean inputs. In this paper, we propose a straightforward data poisoning method targeting the BERT architecture. Our approach does not involve complex modifications to the model or its training process; instead, it relies solely on altering a small portion of the training data. By introducing simple perturbations into just 10% of the training dataset, we demonstrate the feasibility of injecting a backdoor into the model. Our experimental results show a high attack success rate, indicating that the model trained on the poisoned data can reliably associate the trigger with the attacker's desired outputs while its performance on clean data remains unaffected. This highlights the stealth and effectiveness of our method, emphasizing the need for improved defensive strategies to protect against such threats. Our study underscores the critical importance of ongoing research and development to safeguard AI systems from malicious exploitation, ensuring the security and reliability of NLP applications.

This project investigates the behavior of multi-head attention in Transformer models, specifically focusing on the differences between benign and trojan models in the context of sentiment analysis. Trojan attacks cause models to perform normally on clean inputs but exhibit misclassifications when presented with inputs containing predefined triggers. We characterize attention head functions in trojan and benign models, identifying specific 'trojan' heads and analyzing their behavior.

The combination of e-commerce and the green economy allows enterprises to distribute their green products directly to consumers through Internet channels. This paper investigates the impact of service spillovers on emission reduction and channel decisions in a two-echelon supply chain,in which the supplier is a Stackelberg leader and the retailer is a follower.By comparing the equilibrium outcomes of the single-channel and dual-channel supply chain in a setting with and without retail services, we observe that the supplier prefers to encroach on the market when services that retail locations provide largely spillover to and benefit the direct sales channel. Contrary to popular belief, a higher degree of service spillovers is beneficial for the retailer to get more returns under the dual-channel structure, whereas supplier encroachment will lead to a decline in the service level if the spillover degree is relatively low. In addition, the emission reduction level of products under supplier encroachment is always higher than that employed in the single-channel structure if consumers have both low-carbon preference and high degree of service sensitivity. Furthermore, we extend the equilibrium decisions under a carbon cap-and-trade (CCT) mechanism; the results fully verify the above conclusions. Interestingly, the implementation of a carbon price not only motivates carbon emission abatement and supplier encroachment, but also helps achieve a win-win outcome for supply chain members when service spillovers happen.