Paper reading (2025)

Copied from: Public/Paper Reading 2024

このセミナーについて (About this seminar)

担当者が見つけた面白い研究を紹介するセミナーです．以下の内容で構成されます． In this seminar, presenters share interesting studies that they found. The seminar consists of the following elements.

  1. 論文紹介 (Paper presentation): 面白いと思う論文（基本的にLong Paper）を選んで，その内容を著者の代わりになったつもりで発表する． A presenter reads a paper (long paper is strongly preffered) that he or she finds interesting, and explains the content as if he or she was an author of the paper.
  2. 著者プレゼンの上映と解説 (Watching talk videos): 担当者は面白いと思う研究の発表動画を選び，その論文の内容を把握する．担当者の補足説明を聞きながら，参加者みんなで発表動画を鑑賞する． Finding an interesting video presentation, a presenter reads and understands the content of the paper. Attendees watch the presentation video of the author of the paper. The presenter is expected to explain supplementary information about the research.

前者 (1) はプレゼンテーションの練習を兼ねています．後者 (2) は「よい」プレゼンテーションを鑑賞しながら，英語でのプレゼンテーションやディスカッションに慣れることを狙っています． The former (1) aims at practicing presentations. The latter (2) aims at improving listening and discussion skills in English as we enjoy good presentations.

発表者は一人目は (1) を，二人目は (1) か (2) のどちらかを担当します． The 1st presentator take (1). The 2nd one can select (1) or (2).

座長(chair)は、セミナーの司会進行をおこないます。 chair person host the seminar. 座長の仕事を参考にして，円滑に議論が進むように心がけてください． Please contribute to active discussion refering to Chair’s Job.

発表時間 (Presentation Time)

  • Presentation: 15 ~ 20 minnutes
  • QA: 5 ~ 10 minutes

日時 (Date and time)

  • 3Q : 8:50~ (Thu.)

参加者 (Attendee)

  • 全員

:exclamation: 発表を登録するときは，著者名，発表年，タイトル，会議名／ジャーナル名，（巻，号，ページ番号など）を必ず記入してください．preprintの論文は発表しないでください。 :exclamation: Please include author names, publication year, title, conference/journal name, (volume and page numbers) when you add an entry for presentation. Do not introduce papers of preprints.

Tips

  • リモートミーティングでビデオプレゼンテーションを行う場合は，画面およびオーディオを共有してください．
  • When you stream the video presentation via remote meeting, please share the both screen and system audio.
    • For Zoom: You don’t have to do nothing special. Just share your screen.
    • For Microsoft Teams: Please activate Share the system audio feature.

今後の予定 (Planned Seminars)

2025/07/10 8:50-10:30

  • seminar
  • chair
    • Takahashi

      Past Seminar

      2025/04/17 8:50-10:30

  • Clean-Up
  • seminar
    • koike
      • How Far Can We Extract Diverse Perspectives from Large Language Models?
        • Shirley Anugrah Hayati, Minhwa Lee, Dheeraj Rajagopal, Dongyeop Kang; EMNLP2024 main
        • Paper Slide
    • muraoka
      • Zheng et al., Iterated Learning Improves Compositionality in Large Vision-Language Models. CVPR. 2024. [paper] [video] [slides]
  • chair
    • shimada

2025/04/24 8:50-10:30

  • seminar
    • Saito
      • Zeng et al. “Token-level Direct Preference Optimization.” ICML Spotlight. 2024. [paper] [slides]
    • LUO (arase lab)
  • chair
    • David

2025/05/01 8:50-10:30

火曜授業扱いのため休講

2025/05/08 8:50-10:30

  • seminar
    • Katsumata
      • Wynter et al. “RTP-LX: Can LLMs Evaluate Toxicity in Multilingual Scenarios?” AAAI2025. [paper] [slides]
    • Shiotani
      • Ye et al. “Effective Large Language Model Adaptation for Improved Grounding and Citation Generation” NAACL. 2024. [paper][slides]
  • chair
    • Yamada

2025/05/15 8:50-10:30

  • seminar
    • David
    • Onami
      • Liao et al. “DocLayLLM: An Efficient Multi-modal Extension of Large Language Models for Text-rich Document Understanding” CVPR2025 [paper] [slides]
      • I’m out during 9:10-9:20 so I might be a little late to attend.
  • chair
    • Ishikura

2025/05/26 17:15-18:55 ※Pay attention to the start time.

  • seminar
    • Maeda
      • OmniCorpus: A Unified Multimodal Corpus of 10 Billion-Level Images Interleaved with Text (Li+, ICLR 2025 Spotlight Paper)
      • arXiv: https://arxiv.org/pdf/2406.08418, OpenReview: https://openreview.net/forum?id=kwqhn2VuG4
      • 20250526_maeda_omnicorpus.pdf (1.7 MB)
    • Takahashi
      • A Probabilistic Perspective on Unlearning and Alignment for Large Language Models
      • Scholten et al. ICLR 2025 oral
      • paper, slides
  • chair
    • Saito

2025/05/29 8:50-10:30

  • seminar
    • Shimada
      • Qi et al., Safety Alignment Should Be Made More Than Just a Few Tokens Deep,
      • ICLR 2025 oral （Outstanding Papers）
      • paper, slide
    • Yamada (Arase lab)
  • chair
    • Tamura (Arase lab)

2025/06/05 8:50-10:30

  • seminar
    • Han: Align Sentence Simplification with ESL Learner’s Proficiency for Language Acquition
      • Guanlin Li, Yuki Arase, Noel Crespi, NAACL 2025
    • Ishikura (Arase lab)
  • chair
    • Ito (Arase lab)

2025/06/12 8:50-10:30

  • seminar
    • Kobayashi (Arase lab)
    • Tamura (Arase lab)
  • chair
    • Han (Arase lab)

2025/06/19 8:50-10:30

  • seminar
    • Ito (Arase lab)
    • Hida
  • chair
    • Kobayashi (Arase lab)

2025/06/26 8:50-10:30

  • seminar
    • Ma
      • Training Language Models to Self-Correct via Reinforcement Learning
        • Aviral Kumar · Vincent Zhuang · Rishabh Agarwal · Yi Su · JD Co-Reyes · Avi Singh · Kate Baumli · Shariq Iqbal · Colton Bishop · Rebecca Roelofs · Lei Zhang · Kay McKinney · Disha Shrivastava · Cosmin Paduraru · George Tucker · Doina Precup · Feryal Behbahani · Aleksandra Faust, ICLR 2025 Oral
        • paper, video
    • Oba
      • Large Language Diffusion Models
        • Shen Nie, Fengqi Zhu, Zebin You, Xiaolu Zhang, Jingyang Ou, Jun Hu, Jun Zhou, Yankai Lin, Ji-Rong Wen, Chongxuan Li, Arxiv Feb 2025
        • paper, slide
  • chair
    • Matsushita

2025/07/03 8:50-10:30

  • seminar
    • Ichinose
      • DiLM: Distilling Dataset into Language Model for Text-level Dataset Distillation
        • Aru Maekawa, Satoshi Kosugi, Kotaro Funakoshi, and Manabu Okumura, NAACL 2024
        • paper, slide
    • Matsushita
      • Language-Specific Neurons: The Key to Multilingual Capabilities in Large Language Models (Tang et al., ACL 2024) [paper] [slides]
  • chair
    • Shimada