编程 / 框架Programming
- Python
- C++
- Mathematica
- SQL
- PyTorch
- TensorFlow
徐北辰Beichen Xu Beichen Xu徐北辰
非雨（ThisRainIsNotARealRain）创始人 / CEO Founder & CEO, ThisRainIsNotARealRain
公司业务为二次元实体与商业化出版，以本体驱动的连续出版为核心方法。此前在 AI4Science 与计算机视觉方向从事研究与工程。 The company works in anime physical goods and commercial publishing, built on ontology-driven continuous publishing. Before this I worked in AI4Science and computer vision, in both research and engineering.
- 现在Now
- 非雨 · 二次元实体与商业化出版ThisRainIsNotARealRain · anime goods & publishing
- 此前Before
- AI4Science · 计算机视觉与多模态AI4Science · computer vision & multimodal
- 教育Education
- 帝国理工 MRes · 伯明翰 BSc（一等）Imperial MRes · Birmingham BSc (First)
- 位置Based in
- 北京海淀 · 开放合作Beijing · open to collaborate
01 / About
关于About
我是非雨（ThisRainIsNotARealRain）的创始人兼 CEO。公司业务为二次元实体与商业化出版，以本体驱动的连续出版组织 IP 的开发节奏。此前的研究与工程背景在计算机视觉与多模态、AI4Science 与 Physics ML。 I’m the Founder and CEO of ThisRainIsNotARealRain. The company works in anime physical goods and commercial publishing, using ontology-driven continuous publishing to structure how an IP is developed. My research and engineering background is in computer vision and multimodal learning, AI4Science and physics ML.
方向Focus
二次元实体与商业化出版 / AIGC 系统 / 计算机视觉与多模态Anime goods & commercial publishing / AIGC systems / computer vision & multimodal
能力Strengths
模型研发、数据融合、可复现训练与评估、多GPU工作流Modeling, data fusion, reproducible training & evaluation, multi-GPU workflows
合作Collaboration
跨学科协作，把原型推进到可交付组件Cross-functional work from prototype to deployable components
- Transformers
- 扩散模型Diffusion
- GNN
- LoRA
- PyTorch
- AI4Science
02 / Experience
经历Experience
-
— 至今Present
创始人 / CEOFounder / CEO
非雨 ThisRainIsNotARealRain
- 二次元实体：覆盖从设计落地到生产与履约的全部环节，以实际销量评估 IP 的商业潜力。 Anime physical goods: the full chain from product design through production and fulfillment, with actual sales assessing an IP’s commercial potential.
- 商业化出版：经验证的 IP 推进至品牌授权，按品类与地区分别签约，构成公司主要收入。 Commercial publishing: validated IP moves into brand licensing, signed separately by category and region — the company’s primary revenue line.
- 自研 RainStudio 与 RainHub，分别支撑内容开发与创作者社区、周边履约；设定与授权记录集中管理，新项目可直接复用。 Built RainStudio and RainHub in-house, covering content development and the creator community and merch fulfillment respectively, with settings and licensing records managed centrally for reuse across projects.
-
—
AI 算法工程师AI Algorithm Engineer
MapTech · 北京Beijing
- 计算机视觉/视觉语言方向：构建 VL 与多模态学习流程，面向 GIS 与遥感任务（道路提取、轨迹建模、卫星影像理解）。Computer vision & vision–language: built VL/multimodal pipelines for GIS and remote sensing (road extraction, trajectory modeling, satellite understanding).
- 主导全参微调与 LoRA 轻量适配实验（如 Qwen-VL），提升跨区域/跨模态数据集效率。Led full fine-tuning and LoRA adaptation experiments (e.g., Qwen-VL) to improve efficiency across heterogeneous geospatial datasets.
- 设计地理特征 tokenization 与融合策略（patch、轨迹 embedding、hint maps），支撑分割与推理任务。Designed geospatial feature tokenization and fusion (image patches, trajectory embeddings, hint maps) for segmentation and reasoning.
- 搭建可复现的多GPU训练/评估工作流，优化数据标注对齐与 benchmarking。Built reproducible multi-GPU training/evaluation workflows and improved curation, alignment, and benchmarking.
- 与产品/工程团队合作，将研究原型落地为生产 pipeline 组件。Collaborated with product/engineering to translate prototypes into deployable components.
-
AI 算法研究员AI Algorithm Researcher
Science42.tech · 北京Beijing
- 研究 Transformers / Diffusion / GNN 在流体、医学与能源等科学场景的应用。Researched Transformers/Diffusion/GNNs for scientific domains including fluids, biomedicine, and energy systems.
- 与清华大学、西湖大学、IMEC（比利时）等团队保持学术合作与交流。Maintained close collaborations with Tsinghua, Westlake University, and IMEC (Belgium) through joint studies and exchanges.
- 开发 physics-informed AI 模型并推动高质量论文产出（投稿至《Physics of Fluids》等）。Developed physics-informed AI models and contributed to high-quality publications (e.g., submissions to Physics of Fluids).
- 与跨学科团队协作研究代码与流程，并指导实习生完成科研与技术任务。Worked with interdisciplinary teams on research pipelines and mentored interns on technical/research tasks.
-
暑期实习Summer Intern
Kavli Institute for Cosmology · 剑桥Cambridge
- 研究 ML 增强的 Bayesian inference：使用 nested sampling 训练宇宙学数据，并用 masked aggressive flow 生成后验。Worked on ML-enhanced Bayesian inference for cosmology: nested sampling + masked aggressive flow for posterior generation.
- 参与开源 Python 包 “margarine” 的贡献与协作。Contributed to the “margarine” Python package on GitHub.
-
暑期实习Summer Intern
紫金山天文台 · 南京Purple Mountain Observatory · Nanjing
- 对 GRB 现象做系统性综述；用 Python 对 Swift GRB 数据进行拟合与分析。Reviewed GRB literature and analyzed Swift GRB data with Python (fitting and statistical analysis).
- 使用统计方法改进参数估计，熟悉 iminuit 的最小化与误差分析流程。Applied statistical methods for parameter estimation; used iminuit for minimization and uncertainty analysis.
03 / Projects
项目Projects
AeroDiT（用于 RANS 的扩散 Transformer） AeroDiT (Diffusion Transformers for RANS)
将扩散模型与 Transformer 结合，用于 Reynolds-Averaged Navier–Stokes 空气动力学仿真（翼型流动）。 Diffusion Transformers for Reynolds-Averaged Navier–Stokes simulations of airfoil flows.
- 扩散模型Diffusion
- Transformers
- 流体力学Fluid Dynamics
MRes 项目 · 引力波 WGAN MRes Project · Gravitational Wave WGAN
模拟多源引力波时序信号，训练 WGAN 以从信号中重建参数；使用 HPC 进行训练与可视化分析。 Simulated multi-source gravitational-wave time series and trained WGANs to reconstruct parameters; trained and analyzed on HPC.
- PyTorch
- PyTorch Lightning
- 高性能计算HPC
多模态 GIS / 遥感训练流程 Multimodal GIS / Remote Sensing Pipelines
面向大规模地理遥感任务（道路提取、轨迹建模、卫星影像理解）构建 VL/多模态训练与评估流程，包含 LoRA 与全参微调实验。 Built vision–language / multimodal pipelines for large-scale GIS and remote sensing tasks (road extraction, trajectory modeling, satellite understanding), including LoRA and full fine-tuning experiments.
- 视觉语言Vision–Language
- LoRA
- 多 GPUMulti-GPU
机器学习增强的贝叶斯推断（嵌套采样） ML-enhanced Bayesian Inference (Nested Sampling)
在宇宙学推断中结合嵌套采样与机器学习（masked aggressive flow），用于后验分布生成与参数估计；并参与维护开源 Python 包。 ML-enhanced Bayesian inference for cosmology: nested sampling + masked aggressive flow for posterior generation and parameter inference; contributed to an open-source Python package.
- 贝叶斯Bayesian
- 嵌套采样Nested Sampling
- Python
04 / Education
教育Education
MRes
2023 — 2024
帝国理工学院 Imperial College London
Machine Learning and Big Data in the Physical Sciences · 物理系 Machine Learning and Big Data in the Physical Sciences · Department of Physics
- 时间序列Time Series
- 加速数据处理Accelerated Data Processing
- FPGA
- 宇宙学Cosmology
BSc
2020 — 2023
伯明翰大学 University of Birmingham
Physics · 一等荣誉 Physics · First Class Honours
- 统计学Statistics
- 相对论Relativity
- 量子力学Quantum Mechanics
- 线性代数Linear Algebra
06 / Skills
技能与奖项Skills & Awards
工程工具Tooling
- Linux / Unix
- Bash
- vim
- git
- LaTeX
- Docker
语言Languages
- 中文Chinese母语Native
- 英文English流利Fluent
- 日语JapaneseJLPT N3
奖项Awards
- 2020南通市长科技创新奖Nantong Mayor’s Award for Science & Technology Innovation
- 2019江苏省物理奥赛一等奖Physics Olympiad (Jiangsu) — First prize
- 2018未来科学家大奖（银奖）Award Program for Future Scientists — Silver
07 / Contact
联系Contact
如需交流合作、项目讨论或学术/工程问题，欢迎通过邮件或社交平台联系我。 For collaboration, project discussions, or research/engineering topics, feel free to reach out via email or social platforms.