学术讲座

大数据管理与分析方法研究北京市重点实验室（BDAI）研究生沙龙由中国人民大学高瓴人工智能智能学院与信息学院联合定期举行，本周BDAI重点实验室研讨会由高瓴人工智能学院博士生周昆和信息学院博士生赵展浩分别介绍各自的研究工作。欢迎同学们积极参与研讨！

报告题目：Refining BERT for Better Robustness and Effectiveness

报告人简介：周昆，博士二年级

导师：赵鑫，文继荣

研究方向：自然语言处理，推荐系统

Abstract：

Recent works have shown that powerful pretrained language models (PLM) can be fooled by small perturbations or intentional attacks. To solve this issue, various data augmentation techniques are proposed to improve the robustness of PLMs. However, it is still challenging to augment semantically relevant examples with sufficient diversity. In this work, we present Virtual Data Augmentation (VDA), a general framework for robustly fine-tuning PLMs. Based on the original token embeddings, we construct a multinomial mixture for augmenting virtual data embeddings, where a masked language model guarantees the semantic relevance and the Gaussian noise provides the augmentation diversity. Furthermore, a regularized training strategy is proposed to balance the two aspects. Extensive experiments on six datasets show that our approach is able to improve the robustness of PLMs and alleviate the performance degradation under adversarial attacks.

报告题目：Transaction Processing in Distributed Databases分布式数据库中的事务处理

报告人简介：赵展浩，博士三年级

导师：杜小勇

研究方向：分布式数据库系统、事务处理

Abstract：

Informally, serializability means that transactions appear to have occurred in some total order. In this paper, we show that only the serializability guarantee with some total order is not enough for many real applications. As a complement, extra partial orders of transactions, like real-time order and program order, need to be introduced. Motivated by this observation, we present a framework that models serializable transactions by adding extra partial orders, namely multi-level serializability models. Following this framework, we propose a novel concurrency control algorithm, called bi-directionally timestamp adjustment (BDTA), to support multi-level serializability models in distributed database systems. Our experiments show the performance gaps among serializability levels and confirm BDTA performs better than state-of-the-art concurrency control algorithms. (本次报告将分享我们在分布式数据库事务处理机制方向的研究工作。通常可串行化被定义为事务的实际执行结果与某一种串行执行结果相同，是数据库中事务执行的正确性法则。然而，分布式数据库由于其架构的特殊性，保证可串行化仍不能满足一些实际应用的要求，这给事务调度的正确性带来了新的挑战。我们考虑在事务调度中引入额外的一致性偏序要求（如实时序和程序序等），提出了一个多级可串行化模型，该模型通过添加额外的偏序来对分布式数据库中的正确性机制进行理论建模。然后，遵循这个框架，我们提出了一种新的并发控制算法，称为双向时间戳调整（BDTA），使得分布式数据库系统可以支持多级可序列化模型。我们的实验表明了方法在局域网和广域网下的使用性，并确认BDTA的性能优势。)