森下 皓文 Terufumi Morishita
自然言語処理を主たる分野としつつ、周辺分野も含めて、研究をしています:
- 自然言語処理: LLMの推論能力を向上させるための、記号論理学に基づく論理推論コーパスの自動生成
- 機械学習: アンサンブル学習の性能を決定づける根本要因の、理論的な解明
- 経済学 x AI: 多数のLLMエージェントを用いた、経済理論に基づくマクロ経済シミュレーション
会社の実用寄りの研究の傍らで、このような、より基礎的な研究を進めてきています。
大学では、物理学(素粒子・宇宙物理学)を研究していました。
「現象の背後に潜む普遍的な原理」の探求に興味があります。 自然界の原理→物理学、思考の原理→記号論理学(論理推論)、経済現象の原理→経済学、というようにです。
A researcher in AI at Hitachi’s Central Research Laboratory.
I mainly work on natural language processing, including reasoning-capable large language models that help people make better decisions. I also study LLM-agent-based simulations of economic phenomena, such as economic growth and international trade. More broadly, I am interested in general machine learning methods, including ensemble learning.
Earlier, at the University of Tokyo, I studied particle and astroparticle physics at the Kavli Institute for the Physics and Mathematics of the Universe under the supervision of Prof. Hitoshi Murayama.
I enjoy formal academic disciplines such as physics, mathematics, economics, and symbolic logic.
主な研究・発表selected work
-
AAAI (招待講演)Can We Teach Logical Reasoning to LLMs? – An Approach Using Synthetic CorporaIn AAAI Bridge Program "Logical and Symbolic Reasoning in Language Models" keynote speech, 2026
-
招待講演 (最優秀論文賞)
-
YANS2025 (招待講演)
-
IBIS
-
自然言語処理 (最優秀論文賞)
-
人工知能学会 (優秀賞)
-
NeurIPSEnhancing Reasoning Capabilities of LLMs via Principled Synthetic Logic CorpusIn Annual Conference on Neural Information Processing Systems, 2024, 採択率25.8%Acceptance rate 25.8%
-
NLPコロキウム (招待講演)
-
IBIS
-
人工知能学会
-
LREC-COLINGJFLD: A Japanese Benchmark for Deductive Reasoning based on Formal LogicIn Proceedings of the Joint International Conference on Computational Linguistics, Language Resources and Evaluation, 2024, 採択率52%Acceptance rate 52%
-
言語処理学会
-
人工知能学会
-
人工知能学会
-
ICMLLearning Deductive Reasoning from Synthetic Corpus based on Formal LogicIn International Conference on Machine Learning, 2023, 採択率27.9%Acceptance rate 27.9%
-
言語処理学会
-
ICML (Spotlight=上位5%)Rethinking Fano’s Inequality in Ensemble LearningIn International Conference on Machine Learning, 2022, Spotlight (上位5%), 採択率21.9%Acceptance rate 21.9%
-
SemEval (1st prize)Hitachi at SemEval-2020 Task 3: Exploring the representation spaces of transformers for human sense word similarityIn Proceedings of the Fourteenth Workshop on Semantic Evaluation, 2020, 1st prize
-
SemEval (1st prize, Oral)Hitachi at SemEval-2020 task 7: Stacking at scale with heterogeneous language models for humor recognitionIn Proceedings of the Fourteenth Workshop on Semantic Evaluation, 2020, Oral, 1st prize
-
SemEval (2nd prize)Hitachi at SemEval-2020 task 8: Simple but effective modality ensemble for meme emotion recognitionIn Proceedings of the Fourteenth Workshop on Semantic Evaluation, 2020, 2nd prize