■２−１■ 日本データベース学会上林奨励賞を受賞して ～My experience as an industrial researcher～ Seng Pei Liew (LINE株式会社)
It is a great honor for me to be awarded the Kambayashi Young Researcher Award by the Database Society of Japan (DBSJ).
First of all, I would like to express my deepest gratitude to my great mentors and collaborators, including Prof. Masatoshi Yoshikawa, Prof. Yang Cao, Mr. Takagi, Mr. Kato, Mr. Takahashi and Mr. Ueno. I would also like to thank the individual who recommended me for this prestigious award. I am honored to receive recognition for my research on privacy-enhancing technology, particularly differential privacy, which leads to publication in SIGMOD22 and ICLR22. Let me briefly describe my experience that leads to these research projects.
Working as a researcher in industry, as I gained industrial experience in protecting user privacy, I have come to realize that gaining user trust is also important in practical applications. Particularly, it is difficult to persuade a user to send her (raw) data to a curator as this requires user to trust the curator to handle the data securely. This motivated me to study a decentralized architecture of data anonymization where users collaborate among themselves without requiring a centralized entity. This led me to the proposal of network shuffling, of which the work is published in SIGMOD22.
For my ICLR22 work, I was attempting to improve generative modeling with differential privacy. Previous works mainly focused on utilizing DPSGD, a popular method of guaranteeing privacy by adding noise to gradients at each update. I was somewhat unsatisfied with the use of DPSGD, as noise addition at each update erodes the privacy budget, significantly limiting the number of iteration. What came upon me while surveying the literature was a method that preprocess the data with privacy guarantees to train the generative model. Combining this with the idea of using adversarial training led me to a proposal that overcomes the limitations of DPSGD and performs well empirically.
Here are some of my humble thoughts on doing research. I believe I have benefitted from reading a lot (of research papers), and doing critical assessment on them. Furthermore, formulating the right problem to solve has been important to me (sometimes more important than solving the problem itself). In particular, my ideas of formulating problems have come from reading as well as industrial experience. Finally, one must not forget to have fun and enjoy the process of doing research. I hope these thoughts are somewhat useful to up-and-coming researchers. With this I conclude with: have fun and do great research!