Leading AI Companies and Passionate Students Come Together at the 2023 SNU AIIS Spring Retreat!

On May 19, the 2023 Seoul National University Artificial Intelligence Institute (AIIS) Spring Retreat was held in the auditorium and outdoor space of the Seoul National University Engineer House. The Seoul National University AIIS Retreat held every spring and fall is an event at which top-class researchers from Seoul National University and leading companies in the AI field come together to share the latest research trends and interact with one another. At this Spring Retreat, not only did LG AI Research participate, but so did leading AI companies such as CJ, Naver, and Megazone Cloud, all presenting their technological prowess and vision through various programs such as oral sessions, poster sessions, and more.

Hear about the Retreat, at which corporate researchers and students discussed various technology research, and about the members of LG AI Research who attended.

“We hope to work with talented people”

The 2023 Seoul National University AIIS Spring Retreat began with the introduction of companies that have collaborated with AIIS. First, Woohyung Lim, Applied AI Group Leader at LG AI Research, took the stage to present the research areas and achievements of LG AI Research.

“In the short period of two and a half years since its establishment, LG AI Research has published a large number of papers and has steadily improved the quality of such. What’s more, we’re making results by not stopping at only publications, but also making connections to solve difficult problems that couldn’t be solved in existing businesses.”

Lim ended the introduction of LG AI Research with, “LG AI Research is always looking to hire excellent talent in various fields of research. We look forward to working together with talented people at LG AI Research.”

In-depth conversations at poster session

For the Oral Session and Poster Session, three researchers from LG AI Research, Sungmoon Ko, Hyungjun Lim, and Hanseul Jeong, attended. In particular, at the Poster Session, conversations that began as students’ questions deepened and led to meaningful discussions on technology. We had a chat with Sungmoon Ko, Hyungjun Lim, and Hanseul Jeong, who were present at the event.

[AAAI 2023] “Grouping-matrix based Graph Pooling with Adaptive Number of Clusters” – Materials Intelligence Lab Sungmoon Ko (Link)

“It was my first time attending the Seoul National University AIIS Retreat event, and I think it was very meaningful experience particularly because I have a Ph.D. in Physics. Students are able to get information on what kind of research companies are conducting, and see whether the direction of research of their major and the company are aligned.”

In the “Grouping-matrix based Graph Pooling with Adaptive Number of Clusters” by Sungmoon Ko, he suggests a method to overcome information transfer limitations using graph convolution when using a Graph Neural Network. For all node pairs, binary classification was performed to create a “Grouping Matrix” and finally, a method to reduce the graph by applying it to the input node and adjacency matrix was developed. This “GMPool” has differentiated features from existing pooling models in that the model automatically determines the number of groups to be pooled, and as a result, it performed better than the existing model in most of the data and settings tested.

[ICASSP 2023] “Lightweight feature encoder for wake-up word detection based on self-supervised speech representation” – Language Lab Hyungjun Lim (Link)

“After finishing my Ph.D., I continued to develop my career in several companies. Because I spent so much time in university, I thought I had no regrets about the time I spent on campus or in the lab, but it turns out this wasn’t the case. Meeting with such passionate students today and answering questions about their research and career gave me a lot of energy, and it also made me look back on the path I’ve walked on.”

“Lightweight feature encoder for wake-up word detection based on self-supervised speech representation” by Hyungjun Lim is a study aimed at overcoming the limitations of wake-up word detection technology, which requires learning limited-size models with insufficient training data. For this, a compressed self-supervised speech representation was created using an autoencoder-based resource reduction and knowledge distillation technique and applied to various previously proposed models, and a significant performance improvement with minimal parameter increases was observed.

[Neurips Workshop 2022] “ReSPack : A Large-Scale Rectilinear Steiner Tree Packing Data Generator and Benchmark” – Data Intelligence Lab Hanseul Jeong (Link)

“I was surprised by the high level of research topics and achievements of the students, and it motivated me to think, ‘Wow, I need to work harder too’. This gave me the chance to think more about the research topic I was going to work on and gain the energy to deeply immerse myself in it.”

To promote research in the field of STPP (Steiner Tree Packing Problem), Hanseul Jeong presented a generator capable of generating the problem and its benchmark. This was named “ReSPack: A Large-Scale Rectilinear Steiner Tree Packing Data Generator and Benchmark.” STPP refers to the problems related to modeling the components and wiring of a PCB, and compared to other combinatorial optimization problems, this has been a less studied field in AI. However, with ReSPack, problems with feasible solutions up to large sizes of 1024X1024 and constrained STPP introducing constraints used in real domains can be provided. ReSPack is expected to be highly utilized in research and the industry.

Bringing students and researchers closer, the LG Booth

This year's Retreat garnered more interest from students than any other year. Yumi Cho, from Talent Relations Team, who ran the LG AI Research recruitment briefing booth at the event and met students in person, commented that “Many people came to visit us as soon as we started in the morning, so we were really surprised. LG AI Research brochure and goods we had prepared wasn’t enough,” and that “I was proud and felt great because the students seemed really interested in LG AI Research.”

At this event, LG AI Research introduced papers accepted in world-class academic conferences to publicize its research capabilities and vision and confirmed the skills and passion of future AI talents through conversations with them. LG AI Research plans to meet more AI talents who will lead Korea by continuing to participate in industry-university exchanges like this Seoul National University AIIS Retreat. We look forward to meeting with AI talents in the near future.