Samil PwC AI Technology Paper Accepted at World's Leading NLP Conference

First Korean Accounting Firm to Achieve This

AI Autonomously Resolves Unnecessary 'Data Bias'

Samil PwC's proprietary domain-specialized artificial intelligence (AI) representation learning technology has been accepted as a paper at the world's most prestigious conference in the field of natural language processing (NLP).

Samil PwC announced on the 28th that its paper, "REZE: Representation Regularization for Domain-adaptive Text Embedding Pre-finetuning," has been accepted for the main conference at "ACL 2026." This marks the first time that a Korean accounting firm has had a paper published at the ACL main conference. The authors are researchers from the Samil PwC AX Node Gen AI Team.

When training with data of different characteristics simultaneously, AI performance may degrade, or the model can become biased toward a specific task. This is known as the "task conflict and task-induced bias" problem. "REZE," developed by the AX Node Gen AI Team, addresses these limitations by introducing an algorithm that enables AI to autonomously control and suppress unnecessary noise and bias arising during the learning process through a technique called representation regularization. This helps AI selectively learn only the core domain knowledge without interference between different data types.

Lee Seunghwan, Leader (Vice President) of the Samil PwC AX Node, stated, "The acceptance at the ACL 2026 main conference is not a one-off achievement, but the result of ongoing research to enable AI to better understand highly specialized data." He added, "Samil PwC will continue to set new standards for AI services dealing with highly specialized knowledge in fields such as accounting, tax, finance, and law, based on our globally recognized core technologies."

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