Flood Sung
AI Researcher & RL Specialist
Member of Technical Staff at Moonshot AI, leading reinforcement learning research towards AGI. Pioneer in few-shot learning, meta-learning, and the innovative Long Chain-of-Thought reasoning.
About Flood Sung
Flood Sung is a distinguished AI researcher and engineer specializing in deep learning, reinforcement learning, meta-learning, and few-shot learning. Currently serving as a Member of Technical Staff and RL Lead at Moonshot AI, he is at the forefront of developing Artificial General Intelligence (AGI).
With over 6,000 citations on Google Scholar, Flood has published numerous high-impact papers at top-tier conferences including CVPR, NeurIPS, and ICLR. His groundbreaking work on "Learning to Compare: Relation Network for Few-Shot Learning" has achieved over 5,575 citations, establishing him as a leading authority in the field.
Beyond research, Flood is a passionate open-source contributor, maintaining the widely acclaimed "Deep-Learning-Papers-Reading-Roadmap" with over 39.1k GitHub stars. He actively shares his insights through blog posts, academic conferences, and technical talks, earning recognition as an "AGI, Metaverse, and Robotics Evangelist."
"AGI is within reach, and the future lies in successfully applying reinforcement learning to increasingly complex scenarios—from AI simulating driving and writing articles to developing apps and publishing scientific papers."
关于Flood Sung
Flood Sung 是一位在人工智能领域，特别是深度学习、强化学习、元学习和少样本学习方向有深入研究和显著成就的学者和工程师。他目前担任Moonshot AI（月之暗面）的技术专家和RL Lead，致力于推动通用人工智能（AGI）的发展。
根据Google Scholar的数据，Flood Sung的学术成果显著，其论文总引用量已超过6,000次。他在CVPR、NeurIPS、ICLR等顶级学术会议发表了多篇高影响力论文，其中"Learning to Compare: Relation Network for Few-Shot Learning"获得了高达5,575次的引用。
作为开源社区的积极贡献者，Flood Sung维护着广受欢迎的"Deep-Learning-Papers-Reading-Roadmap"项目，获得了超过39.1k的GitHub星标。他通过撰写博客文章、参与学术会议和技术分享，积极传播知识和见解，被同行和媒体描述为"AGI、元宇宙及机器人革命布道师"。
"AGI近在眼前，未来的方向是将强化学习成功应用于更复杂的场景，例如让AI模拟驾驶、创作文章、开发APP、发表科研论文等。"
Research Focus
Few-Shot Learning
Pioneering work on relation networks for learning from limited data, with applications in computer vision and pattern recognition.
Reinforcement Learning
Advancing RL algorithms for sample-efficient learning, meta-RL, and their integration with large language models.
AGI Development
Leading research on Long Chain-of-Thought reasoning and building foundation models for artificial general intelligence.
研究方向
少样本学习
在关系网络方面的开创性工作，用于从有限数据中学习， 在计算机视觉和模式识别领域有广泛应用。
强化学习
推进样本高效学习的RL算法， 元强化学习，以及它们与大语言模型的集成。
AGI发展
领导长思维链推理研究， 为人工通用智能构建基础模型。
Representative Publications
Learning to Compare: Relation Network for Few-Shot Learning
F. Sung, Y. Yang, L. Zhang, T. Xiang, P. H. S. Torr, T. M. Hospedales
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018
Contribution: This landmark paper introduces Relation Networks for few-shot learning, significantly improving model generalization with limited training samples. The approach learns to compare samples through an embedding module and a relation module, achieving state-of-the-art results.
Kimi k1.5: Scaling reinforcement learning with llms
K. Team, A. Du, B. Gao, ..., F. Sung (Team collaboration)
arXiv preprint arXiv:2501.12599, 2025
Learning to Learn: Meta-Critic Networks for Sample Efficient Learning
F. Sung, L. Zhang, T. Xiang, T. Hospedales, Y. Yang
arXiv preprint arXiv:1706.09529, 2017
Actor-Critic Sequence Training for Image Captioning
F. Sung, L. Zhang, T. Xiang, T. M. Hospedales, Y. Yang
IEEE International Conference on Computer Vision (ICCV), 2017
代表性论文
Learning to Compare: Relation Network for Few-Shot Learning
Flood Sung, Y. Yang, L. Zhang, T. Xiang, P. H. S. Torr, T. M. Hospedales
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018
主要贡献： 这篇里程碑式的论文提出了少样本学习的关系网络方法，显著提升了模型在少量训练样本下的泛化能力。该方法通过学习嵌入模块和关系模块来比较样本，实现了最先进的性能。
Kimi k1.5: Scaling reinforcement learning with llms
K. Team, A. Du, B. Gao, ..., Flood Sung (团队合作)
arXiv preprint arXiv:2501.12599, 2025
Learning to Learn: Meta-Critic Networks for Sample Efficient Learning
Flood Sung, L. Zhang, T. Xiang, T. Hospedales, Y. Yang
arXiv preprint arXiv:1706.09529, 2017
Actor-Critic Sequence Training for Image Captioning
Flood Sung, L. Zhang, T. Xiang, T. M. Hospedales, Y. Yang
IEEE International Conference on Computer Vision (ICCV), 2017
TACO-RL: Task Aware Prompt Compression Optimization with Reinforcement Learning
Xiaoqing Zhang, Yuhan Liu, Xiuying Chen, Shuo Shang, Rui Yan, Flood Sung
ACL 2025 (Accepted Findings Papers)
Research Projects
Kimi k1.5 Multimodal Reasoning
Moonshot AI, 2025
Leading the development of Kimi k1.5 multimodal reasoning model, focusing on Long Chain-of-Thought (Long-CoT) technology to achieve state-of-the-art performance in complex reasoning tasks.
Deep Learning Papers Roadmap
Open Source, 2018-Present
Curated collection of deep learning papers with reading roadmap, helping researchers systematically master the field. 39.1k+ GitHub stars and widespread adoption in academia.
Long Chain-of-Thought Research
AGI Research, 2024-2025
Pioneering research on Long Chain-of-Thought reasoning, significantly improving LLM performance in complex reasoning tasks through extended reasoning chains and reinforcement learning optimization.
Relation Network Implementation
Open Source, 2018-Present
Official implementation of the CVPR 2018 paper "Learning to Compare: Relation Network for Few-Shot Learning," providing reproducible code and benchmarks for the research community.
研究项目
Kimi k1.5 多模态思考模型
月之暗面, 2025
主导Kimi k1.5多模态思考模型的研发，专注于长思维链（Long-CoT）技术， 在复杂推理任务中实现最先进的性能。
深度学习论文阅读路线图
开源项目, 2018-至今
精选的深度学习论文集合与阅读路线图，帮助研究者系统性地掌握该领域。 39.1k+ GitHub星标，在学术界广泛采用。
关系网络实现
开源项目, 2018-至今
CVPR 2018论文"Learning to Compare: Relation Network for Few-Shot Learning"的官方实现， 为研究社区提供可复现的代码和基准测试。
Blog & Articles
The Thought Process Behind Kimi k1.5
A comprehensive analysis of OpenAI's o1 model breakthrough and Moonshot AI's journey in developing Long-CoT technology. This in-depth article reveals the technical challenges, strategic decisions, and future implications for AGI development.
Behind Kimi k1.5: From Long-CoT Recognition to Model Training
Personal reflections on the development process of Kimi k1.5, from initial recognition of Long-CoT importance to practical training methodologies.
Latest Advances in Deep Learning and Few-Shot Learning
Comprehensive overview of recent developments in deep learning and few-shot learning, exploring practical applications and future directions.
AGI Discussions: The Future of Artificial General Intelligence
Live discussion series exploring the definition, development paths, and potential impacts of AGI on society and ethics.
The Future of AI and AGI
Insights into AI's future development and the challenges and opportunities facing AGI research and implementation.
博客与文章
Kimi k1.5 背后的长长长长长思考
对OpenAI o1模型突破的全面分析，以及月之暗面在开发长思维链技术过程中的旅程。 这篇深度文章揭示了技术挑战、战略决策以及对AGI发展的未来影响。
Open Source Projects
floodsung
AI Researcher & Open Source Contributor
Deep-Learning-Papers-Reading-Roadmap
Curated collection of deep learning papers with systematic reading roadmap. Most starred repository with 39.1k+ stars.
LearningToCompare_FSL
Official implementation of CVPR 2018 paper "Learning to Compare: Relation Network for Few-Shot Learning".
Deep-Reinforcement-Learning
Collection of deep reinforcement learning algorithms implementation for educational and research purposes.
Meta-Learning-Algorithms
Implementation of various meta-learning algorithms for sample-efficient learning tasks.
开源项目
floodsung
AI研究员 & 开源贡献者
LearningToCompare_FSL
CVPR 2018论文"Learning to Compare: Relation Network for Few-Shot Learning"的官方实现。
Career Timeline
Moonshot AI (月之暗面)
Member of Technical Staff & RL Lead
Leading reinforcement learning research team focused on AGI development. Deeply involved in Kimi series models R&D, particularly Kimi k1.5 multimodal reasoning model and Long-CoT technology research.
Inspir AI
Research Manager
Focused on AGI research as Research Manager, leading projects in advanced AI algorithms and machine learning applications.
ByteDance AI Lab
Research Scientist
Conducted research and development in deep learning and reinforcement learning, contributing to cutting-edge AI applications and algorithms.
Inspir AI
Algorithm Expert
Served as Algorithm Expert, developing and optimizing AI algorithms for various applications and use cases.
Independent Researcher (Early Career)
Independent Research
Worked as an independent researcher in artificial intelligence, focusing on deep learning and reinforcement learning. Self-driven research and open-source contributions laid the foundation for my AI career.
职业经历
Moonshot AI (月之暗面)
技术专家 & RL Lead
领导强化学习研究团队，专注于AGI发展。深度参与Kimi系列模型研发， 特别是Kimi k1.5多模态思考模型和长思维链技术研究。
Inspir AI
研究经理
作为研究经理专注于AGI研究，领导先进AI算法和机器学习应用项目。
字节跳动 AI Lab
研究科学家
从事深度学习和强化学习的研究与开发，为前沿AI应用和算法做出贡献。
Inspir AI
算法专家
担任算法专家，为各种应用和用例开发和优化AI算法。
独立研究员（早期职业）
独立研究
作为独立研究员，专注于人工智能领域，重点是深度学习和强化学习。自主研究和开源贡献为我的AI职业奠定了基础。
Get In Touch
Contact Information
For research collaborations, speaking opportunities, or media inquiries, please reach out via email or LinkedIn.
Quick Message
联系方式
联系信息
如有研究合作、演讲机会或媒体采访需求，请通过邮箱或LinkedIn联系。