総合研究会（2023年度）

Copied from: Public/Research Seminar 2022

毎週メンバー全員が参加し，研究について議論を行うミーティングです．

This is a weekly meeting where all members attend and discuss the research in the laboratory.

  • 日時 (Time slot):
    • 1Q: 08:50~ (Fri.)
  • 発表要領 (Presentation format):
    • （ひとりあたり）発表15 ~ 20分 および 議論10 ~ 15分，合計30分を目安にしてください．
    • 15 - 20 min for a presentation and 10 - 15 min for discussion, total 30 min.
    • Tipsの発表は5分程度（長くても質疑含めて15分くらい）が望ましいです
    • 5 mins for Tips presentation, upto 15 mins including discussion.
  • 言語 (Language used for presentation):
    • トークとスライドのうち，少なくともどちらか一方は英語で用意してください．
      学部生は例外的に，本人が希望すれば，トークとスライドの両方を日本語にすることができます．
    • Please prepare, between talk and slide, at least one of them must be English.
      Undergraduate students are exceptional: both talk and slide can be Japanese if they prefer it.
  • 座長 (Chair)
    • B/Mの学生は持ち回りで座長を担当します．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．
    • B/M students are assigned as a chair of the seminar. Please contribute to active discussion refering to Chair’s Job.

Planned Seminars

2024-03/01 (Fri) 10:45-

  • seminar
    • endo
    • mizuki-kentaro
  • tips
    • ma
  • chair
    • kato

      2024-03-04 (Mon) 13:00~

      Rehearsal for ANLP 2024

  • oral: 15 mins presentation + 5 mins comments
  • poster: 10 mins presentation + 5 mins comments

| Time | Presenter | Form | Slide | | — | — | — | — | | 13:00 - 13:20 | Ohi | oral || | 13:20 - 13:40 | Wata | oral || | 13:40 - 13:55 | Ma | poster || | | | | | | 14:05 - 14:20 | Yang| poster || | 14:20 - 14:40 | mizuki iida | oral | | | 14:40 - 15:00 | Koike | oral | | | | | | | | 15:10 - 15:25 | Koyama | poster || | 15:25 - 15:40 | Maeda | poster || | 15:40 - 16:00 | Loem | oral | | | | | | |

waiting list (cancels)

Please be prepared to do your presentation anytime.

  • sangwhan
  • yang

Past Seminars

2023-04-14 (Fri) 08:50-

  • self introduction edit->Attach Filesを用いてslideをUPして下さい。
    • 後藤 昂彦 (Akihiko Goto)研究生
    • 肥田 礼夢 (Remu Hida) D1（社会人)
    • 水木 健太郎 (Kentaro Mizuki)M1
    • 加藤 靖久 (Yasuhisa Kato)B4
    • 大井 聖也 (Masanari Ohi)B4
    • 綿 祐貴 (Yuki Wata)B4

2023-04/21 (Fri) 08:50-

  • kickoff
    • okazaki: Okazaki Lab 2023 (2.8 MB)

2023-04/28 (Fri) 08:50-

  • seminar
    • kaneko
    • sangwhan
  • tips
    • yang 2023_04_28_tips.pdf (1.4 MB)
  • chair
    • loem hirai

2023-05/05 (Fri) 08:50-

2023-05/12 (Fri) 15:40-

  • seminar
    • sangwhan slides
    • erick, Research Seminar
  • tips
    • koyama left-handed mouse & Dvorak layout
  • chair
    • maeda

2023-05/19 (Fri) 15:40-

  • Talk about LLM by Yokota-sensei

2023-05/26 (Fri) 15:40-

  • seminar
    • mizuki-sakae: Research idea sharing: Multilingual WSD, Promising direction on LLMs
    • koyama rethinkng n-gram based automatic evaluation metric for GEC
  • tips
    • hattori: How to use OpenAI API
  • chair
    • mizuki-kentaro

2023-06/02 (Fri) 15:40-

  • research seeds sharing seminar
    • mizuki-sakae: 1. Knowledge-based Multilingual WSD, 2. Retrieval-augmented QA
      • NOTE: WSD is already presented in the seminar on 2023-05-26.
    • marco: June2023_Research_Proposals.pdf (47.8 kB)
    • kaneko：https://nlp.esa.io/posts/2437
    • okazaki: 研究計画
  • chair
    • ohi

2023-06/09 (Fri) 15:40-

  • seminar
    • yoshikawa: Identifying and Evaluating Factual Knowledge of Pre-trained Language Models
    • cognetta: Implicit Subword Markers
  • tips
    • panatchakorn: Bing Chat is probably better than ChatGPT for academic search
  • chair
    • ishikawa

2023-06/16 (Fri) 15:40-

  • seminar
    • ishikawa: slides
    • loem: Exploring Effectiveness of GPT-3 in GEC: A study on Performance and Controllability in Prompt-based Methods
  • tips
    • loem: Prompt Engineering
  • chair
    • nguyen

2023-06/23 (Fri) 15:40-

  • seminar
    • hida: CheckList for Natural Language Generation
    • ma: Constructing Document-Level Relation Extraction Corpora in Japanese
  • tips
    • hirai:就活Tips
  • chair
    • maeda

2023-06/30 (Fri) 15:40-

  • seminar
    • hattori: Japanese Dialogue Summarization Dataset Construction with Multifaceted Evaluation
    • muraoka: Progress in Unified Multimodal Model (2.4 MB)
  • tips
    • koike: Azure OpenAI API
  • chair
    • endo

2023-07/07 (Fri) 15:40-

  • seminar
    • maeda: Recent Progress
    • nguyen: Causal Reasoning through Two Layers of Cognition for Improving Generalization in Visual Question Answering
  • tips
    • sangwhan
  • chair
    • mizuki-kentaro

2023-07/14 (Fri) 15:40-

  • seminar
    • hirai: Improvement of masked language model by Vokenization considering diversity of assigned images
    • koike: OUTFOX🦊: LLM-generated Essay Detection through In-context Learning with Adversarially Generated Examples
  • tips
    • kaneko
  • chair
    • ishikawa

2023-07/21 (Fri) 15:40-

  • seminar
    • panatchakorn: Mitigating Gender Bias in LLMs by Chain-of-Thought Fine-tuning
    • wang Research_Seminar_0721
  • tips
  • chair
    • loem hattori

2023-07/28 (Fri) 15:40-

  • seminar
    • yang 2023_07_28.pdf (1.9 MB) (LLM as zeroshot manga translator)
    • ao Program Verification Pretraining for Table Fact Verification
  • tips
  • chair
    • hattori loem

2023-08/04 (Fri) 15:40-

  • seminar

    • iida Unsuprevised domain adaptation for searh model using pretrained language model & LLM + IR for QA

    • vijay - A better benchmark for Acceptability evaluation?

  • tips
    • erick
  • self introduction
    • junyoung intro230804.pdf (2.7 MB)
  • chair
    • kato boost

2023-09/22 (Fri) 15:40-

  • seminar

    • endo Japanese Practical Multi-turn Conversation Instruction Tuning Dataset with Human Feedback

    • wata

  • tips
    • hida - Accesibility for Slides
  • chair
    • kentaro-mizuki

2023-09/29 (Fri) 10:00-

  • seminar
    • Ohi: progress.pdf (2.2 MB)
    • mizuki-kentaro
  • tips
    • wang ma: Github: more than source code management
  • chair
    • endo hattori

2023-10/06 (Fri) 15:25-

  • seminar
    • kato Slide: Few-shot inference using English examplesfor multilingual language
    • hidaResearch_Seminar_231006_hida
  • tips
    • ishikawa: Remote - Tunnels on TSUBAME
  • chair
    • ohi

2023-10/13 (Fri) 15:25-

  • seminar

    • koyama N-gram F-score between Source, Reference, and Hypothesis for Grammatical Error Correction

    • yoshikawa Identifying and Evaluating Factual Knowledge of Pre-trained Language Models

  • tips
    • ma wang Tips(Parallelism)
  • chair
    • hattori endo ohi

      2023-10/20 (Fri) 15:25-

  • seminar

    • maeda [slide]

    • ishikawa slides

  • tips
    • muraoka: Remote Debugging (of Python code) with VS Code (1.7 MB)
  • chair
    • ohi endo

2023-10/27 (Fri) 15:25-

School’s festival cleaning

2023-11/03 (Fri) 15:25-

Culture’s day (National Holiday)

2023-11/10 (Fri) 15:25-

  • seminar
    • loem: Enhancing Reasoning with Challenging Teaching Strategy
    • wata: Detecting training data from LLMs without using likelihood
  • tips
    • erick
  • chair
    • hirai

2023-11/17 (Fri) 15:25-

  • seminar
    • ma: Building a Document-Level Relation Extraction Dataset Assisted by Cross-Lingual Projection
    • hattori: Extraction of High-quality Japanese Web Texts from Common Crawl (+ Impressive Points of Google Cloud Next Tokyo ’23)
      • (Tips rather than research)
  • tips
    • yoshikawa: Python packaging
  • chair
    • nguyen

2023-11/24 (Fri) 15:25-

  • 構想発表会リハーサル
    • endo
      • 日本語インストラクションデータセット
    • hattori
      • 多面的な評価を付与した日本語対話要約データセットの構築
    • panatchakorn
      • Mitigating Gender Bias in Large Language Models by Chain-of-Thought Fine-tuning
    • mizuki-kentaro
      • Slide
    • chair
      • maeda

2023-12/01 (Fri) 15:25-

  • seminar
    • muraoka
    • cognetta: Two Counterexamples to Tokenization and the Noiseless Channel
  • tips
    • vijay TIPS FOR PRODUCTIVITY
  • chair
    • panatchakorn loem

2023-12/08 (Fri) 10:45-

  • seminar
    • yang
    • hirai Improvement of masked language model by Vokenization considering diversity of assigned images
  • tips
    • iida Hacking Sentecepiece Model
  • chair
    • kato

2023-12/15 (Fri) 10:45-

  • seminar
    • mizuki-kentaro
    • panatchakorn: Mitigating Gender Bias in LLMs through LLM-generated Feedback
  • tips
    • mizuki-sakae: Introduction to ChatGPT-Assisted research
  • chair
    • wata

2023-12/22 (Fri) 10:45-

  • seminar
    • endo
    • vijay Acceptability Evaluation of Naturally Written Sentences
  • tips
    • marco: Write Short Papers, But First Write Long Papers.pdf (372.5 kB)
  • chair
    • loem panatchakorn

      2023-12/29 (Fri) 10:45-

      2024-01/05 (Fri) 10:45-

  • seminar
    • mizuki-kentaro
    • ohi slide
  • tips
  • chair
    • maeda panatchakorn

      2024-01/12 (Fri) 10:45-

      due to preparation for University Admission Common Test

      2024-01/19 (Fri) 10:45-

  • seminar

    • iida: For Developing Japanese LLM Efficiently~Utilizing Vocabulary Expansion and Bilingual Corpus on Continual Pretraining~

    • koike: How You Prompt Matters! Even Task-Oriented Constraints in Instructions Affect LLM-Generated Text Detection

  • tips
    • yang Model training tips
  • chair
    • panatchakorn maeda

      2024-01/26 (Fri) 10:45-

  • seminar

    • wata: Membership Inference Attacks on LLMs Utilizing Sampling

    • kato

  • tips
    • yoshikawa: Dense vector search with Faiss
  • chair
    • endo

      2024-02/02 (Fri) 10:45-

  • seminar
    • sangwhan
    • nguyen: Causal Reasoning through Two Cognition Layer for Improving Generalization in Visual Question Answering
  • tips
    • maeda: Accelerate your slide production
  • chair
    • wata

      2024-02/09 (Fri) 10:45-

  • seminar

    • Erick: Towards Video presentation translation

    • muraoka: Thoughts on New Multimodal Architectures

  • tips
    • koike: Good Sleep Drives Productivity
  • chair
    • maeda

      2024-02/16 (Fri) 10:45-

  • seminar
    • kato: Multilingual Few-shot learning on low-resoruce-language task
    • yang
  • tips
    • hida: Quick Introduction to Claude
  • chair
    • ohi