R-CCS Cafe 第280回（2025年3月14日）

トップページイベント・広報 R-CCS Cafe R-CCS Cafe 第280回（2025年3月14日）

詳細
開催日	2025年3月14日(金)
開催時間	15:00 - 17:00（15:00 - 16:20　講演者4名による講演、16:20 - 17:00　自由討論（参加自由））
開催都市	兵庫県神戸市／オンライン
場所	計算科学研究センター（R-CCS）6階講堂／Zoomによる遠隔セミナー R-CCS外部の方で参加希望の場合は r-ccs-cafe[at]ml.riken.jp までご連絡ください。
使用言語	発表・スライド共に英語
登壇者	幸朋矢量子HPCプラットフォーム運用技術ユニット技師 Joao Batista 高性能人工知能システム研究チーム特別研究員 Rakesh Teja Konduru データ同化研究チーム特別研究員 Ravi Teja Ponnaganti 量子系物質科学研究チーム特別研究員

講演題目・要旨

1st Speaker: 幸朋矢

Title:
Current Progress in the Quantum-HPC Hybrid Platform Environment Setup

Abstract:
Quantum computers have made rapid advances recently. From an HPC perspective, they can serve as accelerators for offloading specific tasks. Since quantum computers may become powerful enough for more practical use in the future, it is necessary to explore ways to facilitate cooperation between quantum and classical resources. This includes both perspectives on building the integrated platform includes quantum computers and supercomputers and on using them effectively. In this talk, I will outline the design of a multi-site Quantum-HPC Hybrid Platform that is currently under development as the JHPC-quantum project. I will focus on aspects such as resource prioritization, authentication, user-friendly programming evrinonment for seamless access to quantum resources, as well as discuss several challenges encountered during the development of the platform.

2nd Speaker: Joao Batista

Title:
Sentence-level Attribution in Large Language Models

Abstract:
While Large Language Models (LLMs) are being used in a wide range of applications in industry, academia, and by the general population, their use in high-risk applications such as medicine, or law enforcement, among others, is still constrained by the lack of trust and transparency of the models.
Attribution brings trust and transparency to the generated answers by providing a connection between generated text and official sources of information. This allows users to verify whether the generated text matches human-made texts but, it can be expensive from a human-hours point of view depending on the approach taken. Systems like Perplexity and ChatGPT add references to the generated text. However, verifying the references is time-consuming, since the systems provide references to the whole document.
In our current work, we aim to provide a sentence-level attribution system that, besides adding a reference to a document, searches that document for sentences that can back the generated statement. This way, users can verify if the generated sentence has an official source backing it with little effort.

3rd Speaker: Rakesh Teja Konduru

Title:
TBD

Abstract:
TBD

4th Speaker: Ravi Teja Ponnaganti