LG AI Research |

LG AI Research

Advancing AI for a better Life

Together with world-leading AI experts, LG AI Research aims to lead next epoch of AI to realize the promising future with you by providing optimal research environment and leveraging state-of-the-art AI technologies.

Advanced ML

Advanced Machine Learning Lab (AML Lab) is a playground for all-rounders in which cutting-edge AI technology could be researched and developed. We define issues crucial to academics and industry by discovering various data while also innovating the State-Of-The-Art (SOTA) matters of existing problems by designing new models and efficient algorithms. Also we design an efficient system to provide a blueprint for problem resolution.

Our members all aim to possess the DNA which could be categorized into five traits: Innovative, Deep, Effective, Alignment, and Scalable which are namely known as IDEAS. We also respect all creative aspects of engineering including in-depth theories and aim to conduct interaction research where humans and technology can coexist in their evolution while aiming to attain diversity. Through verifying theses and academic research, we demonstrate their actual possibility of realization. We preemptively scale the result of our research by building an open-source library and creating a demo. Also we contribute LG AI Research to achieve performance by making the important deliverables into the project.

Language Processing

Language Processing is a field of research for the understanding and generation of human language and codes composed of text. Language processing technology makes it possible to infer large-scale expertise and provide people with insight, as well as to assist people in their business processes and evolve together by interacting with computers. Currently, we are focusing on detailed fields related to Large Language Models, Code Generation, Question Answering and Reasoning, etc., and have been presenting omnidirectional results in both academic and industrial fields.

Large Language Models

Code Generation and Program Synthesis

Planning, Reasoning, and Acting

Reinforcement Learning

Reinforcement Learning is a research field that searches for the optimal policy that can maximize rewards when given a sequential decision-making problem. Applicable to robotics, natural language processing (NLP), and various real-life problems, it has the advantage of being able to learn even beyond human behavioral policies. Based on the latest reinforcement learning technologies such as Offline RL, RL from Human Feedback, and Transformer-based RL, AML Lab is researching reinforcement learning algorithms that can be applied to various problems in the real world.

Off-policy/Offline RL

RL from Human Feedback and Alignment

Decision/Preference Transformer

Generative Models

Generative Models refer to models that generate virtual data similar to real data and are being researched in various fields such as language, computer vision, molecular structure, and time series data. These generative models are a very important field of research that can be expanded to various applications like chatbots, new drug development, and demand forecasting. At AML Lab, we study generative models theoretically, develop practical methodologies and utilize them in various fields.

Diffusion/Consistency Models

Neural ODE/SDE/RDE

Time-series Forecasting

Foundation and Fundamental

The Transformer Model performs outstandingly in sequential data learning, but requires a lot of resources and data for effective learning, making it difficult to apply directly to data with different geometric structures, such as graphs. To improve this, we are researching to expand the usability and versatility of the Transformer model, such as model efficiency using differential geometry, expansion to non-sequential data, and a methodology for constantly learning and continuously changing data.

Website

https://www.lgresearch.ai/