LG’s ‘EXAONE’ Tops Global Industry-Specialized AI for Tabular and Time Series Data Prediction, Surpassing US and Chinese Big Tech

EXAONE Tabular and Forecast Top Global Leaderboards
Securing Leadership in Industrial Foundation Models

LG's artificial intelligence (AI) platform ‘EXAONE’ has demonstrated its top-tier technological capabilities by surpassing leading big tech companies from the United States and China on globally recognized leaderboards evaluating industrial data prediction performance.

On August 7, LG AI Research Institute announced that ‘EXAONE Tabular,’ which predicts tabular data, and ‘EXAONE Forecast,’ which predicts future data over time, both achieved the world’s highest performance (SOTA—state of the art) in their respective global evaluations. This accomplishment demonstrates LG’s technological competitiveness not just in the arena of general-purpose language models (LLMs), but also in the domain of foundation models for structured data designed to address real-world industrial problems.

On the 19th, at LG Science Park in Magok, Gangseo-gu, Seoul, Baek Jonghoon, Head of LG AI Research Institute, unveiled the super-large multimodal AI 'EXAONE 2.0' during the 'LG AI Talk Concert 2023'. Photo by Hyunmin Kim kimhyun81@

On ‘TabArena,’ a representative global leaderboard for tabular data prediction, ‘EXAONE Tabular’ topped the overall ranking in the categorical data prediction category. Its evaluation score (ELO) was 1,760 points, surpassing the latest model released by Google, ‘TabFM’ (1,749 points). Unlike conventional language models that convert tables into text for analysis, EXAONE Tabular is engineered to directly understand the structure of rows and columns as well as relationships among data points. It maintains high accuracy even in environments with limited or partially missing data (missing values), making it immediately applicable to tasks like manufacturing anomaly detection and quality assessment. LG AI Research Institute plans to conduct proof-of-concept (PoC) projects in three industries—battery manufacturing, bio/healthcare, and finance—during the second half of this year.

In ‘GIFT-eval,’ a leaderboard for time series data prediction, ‘EXAONE Forecast’ claimed first place in the ‘zero-shot’ category, which tests the ability to predict data in domains the AI encounters for the first time without additional training. Having learned from a massive dataset of two trillion simulated real-world time series, this model accurately estimates future changes in product demand, raw material prices, and power consumption. It also ranked second in the ‘Agentic AI’ category, which measures the capability for work automation.

Building on these results, LG AI Research Institute plans to further advance specialized foundation models tailored to specific verticals, such as ‘EXAONE Path’ to support cancer diagnosis, while expanding its range of data handling through the development of ‘Robot Foundation Model (RFM)’ for robotic control.

Woo Hyung Lim, Co-President of LG AI Research Institute, said, “The attention of global big tech companies is rapidly shifting from general-purpose language models to industrial-specialized AI foundation models that solve problems in real-world industrial settings. This achievement proves that the accumulated data from diverse industrial sites can become the source of differentiated competitiveness for Korean AI.”

Soon Young Lee, Head of Data Intelligence Lab at LG AI Research Institute, stated, “LG’s differentiated competitiveness lies in the vast data accumulated across diverse industries and a deep understanding of on-site industrial realities. We will continue to expand our portfolio of foundation models specialized for multiple industrial sectors and establish a new axis of competition in the global AI market.”

This content was produced with the assistance of AI translation services.