コンピューティング科学研究領域のNGUYEN, Minh Le教授が、The 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026) において、SAC Highlightを受賞しました。

　ACL 2026は、自然言語処理および計算言語学分野における主要な国際会議であり、令和8年7月2日～7日に、アメリカ・サンディエゴで開催されました。
　SAC Highlightは、Senior Area Chairs（SACs）が、査読を担当した論文の中から特に優れた論文を選出し、表彰するものです。技術的な新規性、十分な実験的検証、自然言語処理分野への波及効果などが評価されます。

※参考：ACL 2026

■受賞年月日
　令和8年7月7日

■研究題目、論文タイトル等
UniSpec: Training-Free Speculative Decoding for Robust LLM Acceleration Across Languages and Hardware

■研究者、著者
Truong Dinh Do、Nguyen-Khang Le、Le-Minh Nguyen

■受賞対象となった研究の内容
UniSpec is a training-free, lossless speculative decoding framework for accelerating large language model inference across different hardware platforms and languages. It introduces (1) device-aware calibration for optimal draft size selection, (2) confidence-based draft candidate estimation using verifier probabilities, and (3) an improved tree expansion strategy that increases acceptance rates while reducing verification cost. Extensive experiments show consistent inference speedups without compromising generation quality, making UniSpec an effective plug-and-play solution for efficient LLM deployment.

■受賞にあたって一言
We are deeply honored to receive the ACL 2026 SAC Highlight. This recognition is a testament to the dedication, creativity, and collaborative spirit of our research team, and it inspires us to continue pursuing impactful research on efficient, trustworthy, and practical large language models. We are particularly proud that this achievement reflects the exceptional talent and dedication of our young researchers, Truong Dinh Do (completed the JAIST Doctoral Program in September 2026) and Nguyen-Khang Le (JSPS Postdoctoral Research Fellow), whose outstanding contributions were instrumental to the success of this work. This award also highlights Nguyen Lab's commitment to nurturing the next generation of AI researchers and reflects the high-quality research environment at JAIST. We sincerely thank our collaborators, colleagues, and JAIST for their continuous support, which has enabled us to achieve this recognition at ACL, one of the world's premier conferences in natural language processing.

令和8年10月6日