South Korea’s biggest developer conference called DEVIEW, which is held by Naver annually, is set to become a global conference by attracting many top-class AI (Artificial Intelligence) researchers and robotics researchers around the world.
Naver Labs Europe that was acquired by Naver back in 2017 will be at the center of this transition. According to Naver, this year’s DEVIEW will have researchers from Naver Labs Europe participating in a total of five sessions. Naver has decided to place presentations made by the researchers on the first day of DEVIEW 2020 at one place.
In addition to a session related to robots, the researchers will also talk about AI translation technology and performance from their studies on GNN (Graph Neural Network) that has recently started to attract attention within the field of machine learning.
There will be total of four sessions talking about combinations between AI and robot. Gianluca Monash, a researcher from EPFL that is known as the MIT of Europe, will talk about his research experience related to “robot’s reinforcement learning”.
It is difficult to make predictions if robots were to be placed next to people and provide services in various environments. However, modular reinforcement learning method is able to reinforce intelligence by adding new technologies without having to have robots relearn past learnings.
During the session, Monash will divide actions of robot into different learning units through modular reinforcement learning and share an experience that was restructured through a modular robot control architecture.

Photo Image
<A foreign researcher participating in AI for Robotics 2019 is having a conversation with a researcher from Naver Labs.>

A session for vision, which is also called the “eyes of a robot”, is also drawing attention. Naver Labs Europe has proved that its technologies in the field of computer vision are a world-class by participating in many academic competitions around the world. Its deep learning-based algorithm R2D2 that is able to accurately figure out the location through images regardless of environmental changes such as weather, season, time, and lighting is a prime example.
During this session, Naver Labs Europe’s researchers who are in charge of studies on 3D vision will talk about an AI technology “visual localization enhancement method” that is able to make accurate positioning through images. They will also introduce their pipeline “KAPTURE” that combines data sets that are collected through various methods into one format and manages the format efficiently.
A session for meta-learning will also be worthy of attention. Meta-learning is a technique that makes efficient learning possible even with very small number of samples. A researcher from Naver Labs Europe will give a lecture on a method that allows robots to perform tasks effectively even in an environment with many uncertainties. The researcher will share examples that prove that meta-learning is more effective than “one-shot imitation learning” in assisting robots with performing various tasks in a complicated environment.
One Naver official said that some of follow-up studies to a global workshop “AI for Robot”, which involved discussions about potentials of combination between AI and robot with 11 global scholars last year, will be introduced at this year’s DEVIEW.
In addition to sessions involving robots, a top researcher who has more than 1,000 scientific paper citation counts will talk about research experience on GNN that currently receives the highest level of attention in the field of machine learning. In the field of AI translation, Naver Labs Europe will talk about its research result in COVID-19 academic translation model that it has studied along with Naver Papago.
Staff Reporter Kim, Siso | siso@etnews.com