In our team profile articles, we provide a glimpse into our members and company from their perspectives. In this particular article, we highlight Chen Shen, a Senior Research Engineer at Megagon Labs, Mountain View.
Tell us about how you came to Megagon Labs. What attracted you to our company?
In my previous work, I had the opportunity to apply machine learning models and deliver a lot of great products. However, under the regulations and product pressure, we didn’t have much opportunity and time to catch up with novel techniques in the research field. I felt it was time to involve myself in both research and industry. I thought that with a deeper understanding and wider knowledge, I might be able to apply more advanced technology to real-world problems that can be helpful to everyone. I looked into where I might go next and found Megagon Labs to be a good place to accomplish that.
Megagon Labs is different from other research labs. Megagon Labs keeps close collaboration with the Tokyo team, Indeed, and other Recruit subsidiaries. We have rich resources and real-world industry data that can support us as we try to apply technology and research. Most enjoyably, we have more freedom in research. We collaborate with universities and can focus on the most interesting and most valuable parts of our work, resulting in our ability to do more passion-driven work.
What are you currently working on?
In a recent research project, my teammate Jin Wang and I researched long-form text-matching direction. The inputs to the proposed model are long text pairs, for example, long research papers or long stories. Our model could learn effective interaction between the long sequence pairs, and it will tell if the two long docs are matched (for example, if one paper cited another, or if two long paragraphs come from the same long story). It would help real-world downstream applications with long doc matching, like job-to-resume matching, which is a difficult task for BERT-based language models.
In engineering, I’m working on an innovative project called KnowledgeHub, where we create concept knowledge graphs in the HR domain from multiple data sources. Our unique technology sets us apart from the few other companies working on similar projects. The concept knowledge graphs we’ve developed serve as a rich resource and platform for research and applications in knowledge graphs and natural language processing areas. They combine and merge knowledge from multiple data sources and domains, and can easily generate task-specific knowledge graphs based on specific use cases. One exciting aspect is how well-suited our graph is for graph representation learning, making it an ideal candidate for graph learning.
How does working with the teams at Megagon differ from your experiences at other tech companies?
We don’t have a divide between research and engineering. We believe everyone is a researcher. We combine our strengths and have close collaboration where engineers and researchers work side by side.
Additionally, the team encourages us to express our ideas and comments. Even if someone is not directly working on a project, or even if someone’s degree or job title could be different, they are able to share their thoughts and ideas, no matter how critical. Everyone has the opportunity to contribute, and everyone is encouraged to stay open to the perspectives of their colleagues. This allows for more creativity and a greater sharing of everyone’s ideas on our projects. And, because this is a “small” company, we always have access to the leaders of the projects and of the company, and that makes the communication very efficient.
What is something you’ve noticed about Megagon Labs’ development in the time you’ve been here?
If you compare the lab’s output and topics to what it was doing one or two years ago, you will find our topics have changed over time. We have welcomed (and continue to welcome) new people and the fresh knowledge they bring. I really like that. We have the opportunity to bring in new ideas from new people and to find points that could continue the lab’s existing strengths while extending into new areas at the same time.
What piece of advice would you offer to aspiring engineers working in the ML field?
I would be delighted to share two suggestions: Firstly, I encourage us to engage more in communications and to build external connections. Attending conferences and workshops, even without having publications, can be immensely beneficial. By doing so, we stay updated on the latest developments in the field and gain insights into ongoing research. Second, highlight the importance of the connection between research work and real-world applications. Striving to understand how these excellent research findings can be applied to practical problems and positively impact our daily lives, might drive meaningful and impactful contributions to the field.