I am Kosuke Ukita, a PhD student researching AI. I am interested in both practical applications of deep learning and theoretical research in machine learning: - Efficient generative AI and foundation model construction - Machine learning theory and learning dynamics (e.g., Grokking) Recently, I have been focusing on Flow Matching.

News

Selected Publications

1. [1]

   Weight Tying Causes Grokking in Conditional Flow Matching

   Kosuke Ukita, Tsuyoshi Okita

   The 14th International Symposium on Applied Engineering and Sciences (SAES 2026), No. p0092

   November 2026 · Fukuoka, Japan

   Non-refereedOral
2. [2]

   拡散モデルを用いた人間行動モーションデータからの写実的な映像生成

   松石康希, 浮田嵩祐, 大北剛

   情報処理学会論文誌 (IPSJ Journal), 67-10, 2026.

   July 2026 · Japan

   RefereedJournal
3. [3]

   Efficient Inference for Flow Matching via Unified Path CFG

   Kosuke Ukita, Tsuyoshi Okita

   The 30th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 2026)

   June 2026 · Hong Kong, China

   RefereedOral

Grants

• –

  SPRING Scholarship Awardee

  Japan Science and Technology Agency (JST)

  April 2026 - March 2029 · JPY ¥1,800,000 (+ ¥900,000) / year
• –

  Student Travel Award

  The 30th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 2026)

  June 2026 · USD $500
• –

  International Travel Grant

  The Telecommunications Advancement Foundation

  April 2026 · JPY ¥290,000

Awards

• –

  Student Travel Award

  The 30th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 2026)

  June 2026
• –

  Spotlight Presentation (Top 3%)

  The 29th International Conference on Artificial Intelligence and Statistics (AISTATS 2026)

  May 2026
• –

  Promising Researcher Award

  Information Processing Society of Japan (IPSJ) Kyushu Branch

  March 2024

Research Interests

• —Generative AI
• —Deep Learning
• —Machine Learning
• —Foundation Models
• —Multimodal
• —Diffusion
• —Flow