LG Surpasses Google and Alibaba in AI Benchmarks with Proprietary Industrial Data
LG's proprietary industrial data has taken top spots in AI benchmarks, surpassing Google and Alibaba. EXAONE Tabular reached an ELO score of 1,760 on TabArena, outperforming Google's TabFM model with a score of 1,749. This achievement marks a significant milestone for the company, which claims to have developed a more efficient method for analyzing tabular data.
Tabular data represents the dominant type in enterprise and industrial settings, and LG's approach has been proven effective on various tasks such as battery cell quality prediction and disease-risk scoring. The model processes tables in their native format, learning relationships between columns and rows without serializing column names and values as tokens.
The practical consequence of this architecture is that tabular foundation models can generalize to entirely new tables without retraining and outperform gradient-boosted trees on smaller datasets. LG's EXAONE Tabular model has demonstrated this capability, ranking in the top positions on the small-dataset subset of the TabArena benchmark.
LG AI Research has also achieved a significant breakthrough in time-series forecasting with its EXAONE Forecast model. The model claimed first place in the zero-shot category on GIFT-Eval, a rigorous evaluation suite for time-series foundation models developed by Salesforce AI Research. This achievement marks a significant milestone for LG's efforts to develop deployable AI solutions across industrial settings.