[{"data":1,"prerenderedAt":257},["ShallowReactive",2],{"bio-en":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":7,"body":9,"_type":251,"_id":252,"_source":253,"_file":254,"_stem":255,"_extension":256},"/biography/en","biography",false,"","Biography",{"type":10,"children":11,"toc":248},"root",[12,22,55,69,109,114,120,157,180,194,199,238,243],{"type":13,"tag":14,"props":15,"children":18},"element","div",{"className":16},[17],"section-title",[19],{"type":20,"value":21},"text","short",{"type":13,"tag":23,"props":24,"children":25},"p",{},[26,35,37,44,46,53],{"type":13,"tag":27,"props":28,"children":32},"a",{"href":29,"rel":30},"https://kosuke-ukita.github.io/",[31],"nofollow",[33],{"type":20,"value":34},"Kosuke Ukita",{"type":20,"value":36}," is a first-year Ph.D. student at the ",{"type":13,"tag":27,"props":38,"children":41},{"href":39,"rel":40},"https://www.iizuka.kyutech.ac.jp/en/",[31],[42],{"type":20,"value":43},"Graduate School of Computer Science and Systems Engineering, Kyushu Institute of Technology",{"type":20,"value":45},", where they conduct research in the ",{"type":13,"tag":27,"props":47,"children":50},{"href":48,"rel":49},"http://www.alp.ai.kyutech.ac.jp/tsuyoshi/index.html",[31],[51],{"type":20,"value":52},"Okita Laboratory",{"type":20,"value":54},".",{"type":13,"tag":23,"props":56,"children":57},{},[58,60,67],{"type":20,"value":59},"Supported by the ",{"type":13,"tag":27,"props":61,"children":64},{"href":62,"rel":63},"https://www.ccr.kyutech.ac.jp/dc_support/spring",[31],[65],{"type":20,"value":66},"Support for Pioneering Research Initiated by the Next Generation (SPRING)",{"type":20,"value":68}," program, they conduct research on deep learning and generative modeling, with a focus on enhancing the efficiency of Flow Matching and Diffusion Models, as well as exploring the theoretical foundations of representation learning.",{"type":13,"tag":23,"props":70,"children":71},{},[72,74,81,83,90,92,99,101,108],{"type":20,"value":73},"Their recent achievements include a ",{"type":13,"tag":27,"props":75,"children":78},{"href":76,"rel":77},"https://virtual.aistats.org/virtual/2026/spotlight/13338",[31],[79],{"type":20,"value":80},"Spotlight presentation (top 3% of submissions)",{"type":20,"value":82}," at ",{"type":13,"tag":27,"props":84,"children":87},{"href":85,"rel":86},"https://virtual.aistats.org/",[31],[88],{"type":20,"value":89},"AISTATS 2026",{"type":20,"value":91}," and an Oral presentation at ",{"type":13,"tag":27,"props":93,"children":96},{"href":94,"rel":95},"https://www.pakdd2026.org/",[31],[97],{"type":20,"value":98},"PAKDD 2026",{"type":20,"value":100},", where they received the ",{"type":13,"tag":27,"props":102,"children":105},{"href":103,"rel":104},"https://www.pakdd2026.org/awardees#:~:text=Kosuke%20Ukita%20(Kyushu%20Institue%20of%20Technology)",[31],[106],{"type":20,"value":107},"2026 PAKDD Student Travel Award",{"type":20,"value":54},{"type":13,"tag":14,"props":110,"children":113},{"className":111},[112],"py-8",[],{"type":13,"tag":14,"props":115,"children":117},{"className":116},[17],[118],{"type":20,"value":119},"extended",{"type":13,"tag":23,"props":121,"children":122},{},[123,128,130,137,139,146,148,155],{"type":13,"tag":27,"props":124,"children":126},{"href":29,"rel":125},[31],[127],{"type":20,"value":34},{"type":20,"value":129}," was born and raised in Kagawa, Japan. At ",{"type":13,"tag":27,"props":131,"children":134},{"href":132,"rel":133},"https://www.kagawa-edu.jp/kanich02/",[31],[135],{"type":20,"value":136},"Kanonji Daiichi High School",{"type":20,"value":138}," (Science and Mathematics Course), he maintained a top academic standing and received the ",{"type":13,"tag":27,"props":140,"children":143},{"href":141,"rel":142},"https://statedu.jp/cse/sports08.htm#:~:text=%E5%A4%A7%E5%A1%9A%E5%8A%9F%E5%A4%AA%E9%83%8E%E3%83%BB-,%E6%B5%AE%E7%94%B0%E5%B5%A9%E7%A5%90,-%E3%83%BB%E9%96%A2%E3%81%8F%E3%82%8B%E3%81%BF%E3%83%BB%E7%9F%B3%E4%BA%95",[31],[144],{"type":20,"value":145},"Best Award at the Sports Data Analysis Competition",{"type":20,"value":147},", demonstrating an early interest in data-driven approaches. He also participated in a ",{"type":13,"tag":27,"props":149,"children":152},{"href":150,"rel":151},"https://sshkaigai2018.blogspot.com/",[31],[153],{"type":20,"value":154},"research study tour to California, USA",{"type":20,"value":156},", an experience that exposed him to international research environments and set him on a path toward an academic career.",{"type":13,"tag":23,"props":158,"children":159},{},[160,162,169,171,178],{"type":20,"value":161},"In 2020, he enrolled in the Department of Artificial Intelligence, School of Computer Science and Systems Engineering, ",{"type":13,"tag":27,"props":163,"children":166},{"href":164,"rel":165},"https://www.kyutech.ac.jp/",[31],[167],{"type":20,"value":168},"Kyushu Institute of Technology",{"type":20,"value":170},". During his undergraduate years, he focused on deep learning applied to medical imaging, and his Bachelor's thesis proposed a representation-conditioned latent diffusion model for hematoma classification in brain CT images. Concurrently, he joined the research activities of the Okita Laboratory, contributing to work on multi-task self-supervised learning for sensor data (",{"type":13,"tag":27,"props":172,"children":175},{"href":173,"rel":174},"https://dl.acm.org/doi/10.1145/3594739.3610745",[31],[176],{"type":20,"value":177},"UbiComp/ISWC 2023",{"type":20,"value":179},").",{"type":13,"tag":23,"props":181,"children":182},{},[183,185,192],{"type":20,"value":184},"In 2024, he advanced to the Master's Program in the Graduate School of Computer Science and Systems Engineering at the same institution. His Master's research broadened the scope of his work to wearable sensor data. Building on the application of generative models to representation learning, he progressed from an analysis of human activity recognition (HAR) using diffusion models (",{"type":13,"tag":27,"props":186,"children":189},{"href":187,"rel":188},"https://dl.acm.org/doi/10.1145/3675094.3678439",[31],[190],{"type":20,"value":191},"UbiComp 2024",{"type":20,"value":193},") to developing a sensor foundation model trained via joint learning with Flow Matching.",{"type":13,"tag":23,"props":195,"children":196},{},[197],{"type":20,"value":198},"At the center of his research lies the question of how generative modeling and representation learning can be reconciled—and how the learning dynamics underlying both can be elucidated. Rather than treating diffusion models or Flow Matching merely as tools for data generation, he examines the geometric and information-theoretic structure embedded in their score functions and probability flows, seeking to reinterpret these as a theoretical foundation for representation learning. From this perspective, he pursues methods that simultaneously reduce computational cost and improve representation quality.",{"type":13,"tag":23,"props":200,"children":201},{},[202,204,210,211,216,218,223,225,230,232,237],{"type":20,"value":203},"In 2026, he entered the Doctoral Program at the same graduate school. Shortly after, his Master's research was presented as a ",{"type":13,"tag":27,"props":205,"children":207},{"href":76,"rel":206},[31],[208],{"type":20,"value":209},"Spotlight paper (top 3% of submissions)",{"type":20,"value":82},{"type":13,"tag":27,"props":212,"children":214},{"href":85,"rel":213},[31],[215],{"type":20,"value":89},{"type":20,"value":217},". Supported by the ",{"type":13,"tag":27,"props":219,"children":221},{"href":62,"rel":220},[31],[222],{"type":20,"value":66},{"type":20,"value":224}," program, he further advanced the efficiency of Flow Matching inference, presenting the Unified Path CFG at ",{"type":13,"tag":27,"props":226,"children":228},{"href":94,"rel":227},[31],[229],{"type":20,"value":98},{"type":20,"value":231}," as an Oral paper and receiving the ",{"type":13,"tag":27,"props":233,"children":235},{"href":103,"rel":234},[31],[236],{"type":20,"value":107},{"type":20,"value":54},{"type":13,"tag":23,"props":239,"children":240},{},[241],{"type":20,"value":242},"His current research interests have expanded to encompass learning dynamics in deep neural networks. He is particularly drawn to phase transitions in generalization—phenomena such as Grokking, in which a model's generalization performance improves suddenly and dramatically well after training loss has converged. By leveraging conditional generation processes as a means of observing and steering the learning trajectory, he aims to characterize when and why such abrupt structural transitions occur. He believes that the intersection of generative modeling and learning theory harbors a wealth of fundamental, yet largely unexplored, questions.",{"type":13,"tag":23,"props":244,"children":245},{},[246],{"type":20,"value":247},"Outside of research, he enjoys implementing new algorithms and working through mathematical proofs. As the landscape of AI development continues to evolve rapidly and occasions to write code from scratch have become less frequent, he has cultivated the habit of reading highly-regarded open-source projects on GitHub with the same deliberate attention one brings to literature—carefully tracing each design decision. He believes that, much like a well-crafted paper, excellent code encodes the author's thinking and design philosophy.",{"title":7,"searchDepth":249,"depth":249,"links":250},2,[],"markdown","content:biography:en.md","content","biography/en.md","biography/en","md",1791434470644]