Researching the AI of Tomorrow, Sungryull Sohn of LG AI Research US

In a word, he could be described as a model student who is good at both studying and getting along well with people. We have met with researcher Sungryull Sohn, who has come to Korea from LG AI Research US and who likes to dig into areas that he likes. We were able to get a glimpse of his hardened character as a fierce researcher from the way he maintained a bright smile in response to questions about his life as a researcher.


The moment life began as an AI researcher

Sungryull Sohn focused on the AI field early, before there was much public interest in AI. He had a purely academic curiosity in the fact that computers could get so smart.

“I majored in Electrical and Electronic Engineering during my master’s, and I studied a lot about AI because I grew interested in the field. I felt that AI was the future of computers, and finding it very interesting I wanted to research it in earnest. At the time AI was little-known in Korea, and my friends seemed surprised by my interest.

As I studied for my doctorate after that, I began to research in earnest regarding reinforcement learning. I was drawn to the subject after reviewing the paper “Zero-shot task generalization with multi-task deep reinforcement learning” written by a senior in the lab, Dr. Junhyuk Oh. The machine learning that I’d known had a simple structure, predicting output for input. Here, people gave very detailed commands and the AI understood it, carrying them out like humans. I thought it was AI that was truly realistic, and that it was the future.”

Sungryull Sohn, researcher of LG AI Research US

 
As he had much interest in AI and fierce willingness for research, it came as a very natural decision for him to choose to be a researcher. His time as an intern at Microsoft(MS) and Google made him certain that he wanted to be a researcher.

“When I was an MS intern, I didn’t have many teammates. There were 8 researchers together studying reinforcement learning, and we all had the same interest. Even when eating or chatting we talked about the papers we’d recently read, and I loved how we shared ideas. I was able to feel how much synergy can occur when slightly different viewpoints meet.

Because Google was such a large company, I loved the opportunities to meet various types of people. They didn’t always fit in with my field, but there were a lot of connections within a broader framework for these AI experts who specialized in self-driving, industrialization, engineering, and so on. I was able to consider AI that was more closely connected to reality, and I felt how important communication between researchers can be. There were a lot of fun events as well. They weren’t just work-related, and I liked how I could experience well-planned and large internal events such as racing competitions.”


Being an LG AI Researcher, and LG AI Research US

Sungryull Sohn found out about LG AI Research US through his PhD advisor during his doctorate program, Honglak Lee CSAI. Since he wanted to pursue research regarding reinforcement learning, it held much charm.

“I thought about where to go to perform the research that I wanted, and the conclusion was the LG AI Research . There are not many actualization cases in the field of reinforcement learning, and it’s difficult for a company to try. But the LG AI Research had immense interest in reinforcement learning, especially in the embodied agent field where I am hugely interested. I loved the fact that it had a specific vision.”

LG AI Research US is based in the University of Michigan and was launched recently. It is now currently recruiting native researchers actively, and Sungryull Sohn worked there as a researcher. Based on the recommendation of the representative of LG AI Research, , he was able to come to Korea and meet many types of research and researchers.

“After coming to Korea, I’ve been quite busy but having fun. In the morning and evening I have meetings about projects I had going in the US. In the afternoon, I discuss various things with the people here. We mainly talk about future projects and how to select the direction for our research. I’ll be continuing the projects I start here after returning to the US.”

Sungryull Sohn having a online meeting with the LG AI Research US team

Sungryull Sohn is experiencing many things that he was unable to try at LG AI Research US, which is a new organization. His experience with the Korean corporate culture, which he had been away from for a long time, helped him break free from some of his biases.

“I thought that the HR and Culture team are very close. It has been a long time since I worked in a company in Korea, and it was hard to know what efforts were being made in my company for culture and what kind of culture was forming. Right now, Culture3.0 Squad helps us think about our team’s culture and its improvement and helps us out with that. That was very nice.”

Another big advantage he experienced here was “collaboration.” Korean companies have very clear boundaries between teams, and he was of the opinion that collaboration would be difficult. The varied and active collaboration culture in the LG AI Research surprised him very much.

“There’s a Material Informatics(MI) squad within our Fundamental Research(FR) lab. They have a lot of projects about applying basic science such as biology, physics, chemistry, etc. to AI. I thought the FR lab would be focusing only on basic research, but there was a lot of new attempts being made inside the big framework of AI through collaboration. And it’s very appealing that researchers may freely research what they like as long as the minimum amount of reporting is performed.”

With such work environment backing him up, Sungryull Sohn is also carrying out various projects. His main area of research is his principal interest: embodied agents.

“AI is popular these days and is commercialized greatly, but right now we only have online communication services like chatbots or robot cleaners that do not interact with humans at all. We don’t have products or services yet with unique bodies that communicate with people directly. For AI to become really useful, I think they need to reach a point where they can interact perfectly with humans. Embodied agents are what could make that possible. I mainly research the inference abilities of AI.

I also carry out PCB projects with the Data Intelligence Lab. When designing circuits, finding the shortest way to connect the chips is called the routing problem. Experts are solving it themselves at present, but we’re working on a way to use reinforcement learning to figure this out through this project. Once the level reaches that of an expert, LG will be able to make a huge impact. That is immensely appealing.”

He has always done fundamental research only, and applied research that is connected to real life is a very new experience for him. Sungryull Sohn describes the two types of research as having different types of charm.

“Actually, not many companies will provide so much support for the fundamental research areas. Because I have a big goal I want to accomplish through my research, I decided that the LG AI Research was the perfect place to do it. Fundamental research is fun since you can set your own goals, imagine the future, think deeply, and slowly progress.

Applied AI research is about creating strategies and executing them for a closer and surer future. At the very early stages of research, like kick-off meetings, numbers are already available for things like ‘how much cost and time reduction is possible by developing this technology.’ When development succeeds, it can change people’s lives, and that makes us expect a greater sense of accomplishment.”


Expansion and evolution through AI research

Though he is just getting started, Sungryull Sohn is learning a lot at the LG AI Research . His ties to Honglak Lee CSAI, who influenced him greatly, cannot be left out.

“I met him often as the PhD advisor, but we only talked about research. Meeting him in the same company made me nervous about how to treat him, but he made time for me immediately and kept buying me meals. I think he was trying to close the gap so that I would find him approachable. I was really grateful.

And he very accurately points out what I miss in my research. He guides me so that I can think about real-world impact instead of just the research. That’s why I’m able to think more about methods to allow for function in various situations instead of a narrower view.”

Though he will be returning to the US soon, he is enjoying his time (his “Magok life”) in Korea. He has been developing his interests regarding fascinating research areas through discussion with others, and every day he feels like he is on a trip as he hasn’t lived in a large city in a long time.

“As interest in the embodied agent field increased, a large challenge is finally occurring. It’s called BEHAVIOR, and if I can make this time meaningful I’ll be able to get a step closer to the commercialization of reinforcement learning. Personally I’m very excited at the news, and I love to share my interest by talking about it with other researchers.”

Sungryull Sohn, researcher of LG AI Research US

 
This research field has become a huge goal for him as he is passionate about reinforcement learning and embodied agents in particular. He is determined to see it through even if it takes 5 years or 10, and the sense of accomplishment he will feel then is one of his biggest motivations.

“My goal is to produce meaningful research results with the other researchers in the reinforcement learning field. In the longer term, I’d like to take a surer go at projects related to embodied agents. I bought a robot vacuum recently, and I felt how much more comfortable life became with that small change. If robots could do complex tasks like laundry, cooking, and shopping, they’ll bring about a huge change in our lives as much as self-driving cars will. This is a very difficult field of study. As I want complete commercialization, there’s still a long way to go. But interest is growing, and I feel positive about this as meaningful attempts are being made.

Such projects can never be done alone, and I need to collaborate with experts in the reinforcement learning, vision, and language fields to achieve my goals. I want to do that together with the various experts at the LG AI Research . I also wish to become a leader someday that takes the lead in projects. I’d like to develop my teamwork and leadership and other necessary traits so that I can grow in that fashion.”


Check out LG AI Research Recruit (Link)