Few-Shot & Meta-Learning
少样本与元学习
Relation Networks and meta-critic methods for learning from limited supervision and improving transfer.
关系网络与元评论器方法，面向少监督学习与跨任务迁移。
Robotics / RL / Foundation Models
机器人 / 强化学习 / 基础模型
Flood Sung
AI Researcher. Robotics Founder. Agent Systems Builder.
AI 研究员。机器人创业者。智能体系统建设者。
Founder & CEO of XVI Robotics, focused on general-purpose humanoid robot brain systems. Research spans reinforcement learning, meta-learning, few-shot learning, agentic intelligence, and multimodal foundation models.
XVI Robotics 创始人兼 CEO，专注于通用人形机器人的大小脑系统。研究方向覆盖强化学习、元学习、 少样本学习、智能体智能与多模态基础模型。
Affiliation
当前身份
XVI Robotics
Research Identity
研究标识
RL + Agents
11,089
Google Scholar citations
Google Scholar 引用
13
h-index
h-index
45.9K
GitHub stars
GitHub 星标
120
Public repositories
公开仓库
Profile
个人简介
From few-shot learning to embodied intelligence.
从少样本学习到具身智能。
Data refreshed May 12, 2026 from 数据于 2026-05-12 更新，来源： Scholar, GitHub API, OpenAlex
Flood Sung is an AI researcher and engineer working across deep learning, reinforcement learning, meta-learning, few-shot learning, and agentic systems. He currently leads XVI Robotics, building brain systems for general-purpose humanoid robots.
His Google Scholar profile lists 11,089 citations, with research interests in robotics, foundation models, agents, reinforcement learning, and meta learning. His CVPR 2018 paper Learning to Compare: Relation Network for Few-Shot Learning remains a landmark work in few-shot visual recognition.
At Moonshot AI, his recent coauthored work connects long-chain reasoning, reinforcement learning, multimodal models, formal reasoning, and open agentic intelligence through the Kimi series.
Flood Sung 是一位横跨深度学习、强化学习、元学习、少样本学习与智能体系统的 AI 研究员与工程师。 他目前领导 XVI Robotics，打造面向通用人形机器人的大小脑系统。
他的 Google Scholar 主页显示总引用 11,089 次，研究兴趣包括机器人、基础模型、智能体、 强化学习与元学习。CVPR 2018 论文 Learning to Compare: Relation Network for Few-Shot Learning 至今仍是少样本视觉识别方向的代表性工作。
在月之暗面的近期合作研究中，他参与了 Kimi 系列模型相关工作，把长链推理、强化学习、多模态模型、 形式化推理与开放智能体智能连接起来。
Active vectors
当前方向
Robotics
Foundation Models
Agent Systems
Reinforcement Learning
Meta Learning
Multimodal Reasoning
Research System
研究系统
A compact map of the work.
一张紧凑的研究图谱。
The through-line is sample-efficient intelligence: models that reason, adapt, act, and transfer across real environments.
主线是样本高效智能：能够推理、适应、行动，并迁移到真实环境中的模型。
Reinforcement Learning
强化学习
RL for reasoning, planning, formal proof, code generation, and long-horizon agent behavior.
面向推理、规划、形式化证明、代码生成与长程智能体行为的强化学习。
Foundation Agents
基础智能体
Open agentic intelligence, computer-use agents, multimodal models, and tool-using systems.
开放智能体智能、计算机使用智能体、多模态模型与工具使用系统。
Selected Publications
代表性论文
High-impact papers and recent frontier systems.
高影响力论文与前沿系统工作。
Citation counts: Google Scholar, May 12, 2026. 引用数：Google Scholar，2026-05-12。
6,378Citations引用
2,093Citations引用
1,091Citations引用
637Citations引用
149Citations引用
112Citations引用
55Citations引用
181Citations引用
Open Source
开源项目
Research tools that became community infrastructure.
成为社区基础设施的研究工具。
GitHub API snapshot: 120 public repos, 45,931 stars, 8,877 forks. GitHub API 快照：120 个公开仓库，45,931 stars，8,877 forks。
45.9K
stars across public GitHub repositories, led by long-running reading roadmaps and reproducible research code.
公开 GitHub 仓库累计 stars，以长期维护的论文路线图与可复现研究代码为主。
8.9KForks
3.3KFollowers
120Repos
39.5KTop repo stars
Deep-Learning-Papers-Reading-Roadmap
A curated reading roadmap for deep learning papers, widely used by students and researchers.
面向深度学习论文的系统阅读路线图，被学生与研究者广泛使用。
39,4937,306
Meta-Learning-Papers
Paper collection for meta-learning, learning to learn, one-shot learning, and few-shot learning.
元学习、learning to learn、one-shot 与 few-shot learning 论文合集。
2,654474
LearningToCompare_FSL
PyTorch implementation of the CVPR 2018 Relation Network paper.
CVPR 2018 关系网络论文的 PyTorch 官方实现。
1,075264
DRL-FlappyBird
Playing Flappy Bird with deep Q-learning, built as a practical RL learning project.
使用 Deep Q-Learning 玩 Flappy Bird 的强化学习实践项目。
585322
DDPG
Reimplementation of DDPG for continuous control with OpenAI Gym and TensorFlow.
基于 OpenAI Gym 与 TensorFlow 的 DDPG 连续控制复现。
575177
LLM-with-RL-papers
A focused collection of papers on the intersection of LLMs and reinforcement learning.
聚焦 LLM 与强化学习交叉方向的论文合集。
28010
Career Timeline
职业经历
Research, systems, and company building.
研究、系统与公司建设。
A concise public-facing timeline of roles and research programs.
公开主页中的简明经历线索。
2025 - Present
XVI Robotics
Founder & CEO
创始人兼 CEO
Building general-purpose humanoid robot brain systems that integrate perception, reasoning, planning, and control.
打造通用人形机器人大小脑系统，连接感知、推理、规划与控制。
2023 - 2025
Moonshot AI
Member of Technical Staff & RL Lead
技术专家 & RL Lead
Worked on Kimi series research, including long-chain reasoning, RL scaling, multimodal models, and agentic systems.
参与 Kimi 系列研究，包括长链推理、RL scaling、多模态模型与智能体系统。
2022 - 2023
Inspir AI
Research Manager
研究经理
Led AGI-oriented research projects and advanced machine-learning applications.
领导面向 AGI 的研究项目与机器学习应用探索。
2020 - 2022
ByteDance AI Lab
Research Scientist
研究科学家
Conducted research and development in deep learning and reinforcement learning.
从事深度学习与强化学习方向的研究和开发。
Earlier
早期
Independent Research & Open Source
Research foundation
研究基础
Built the research and open-source body of work that led to few-shot learning, RL, and meta-learning contributions.
通过独立研究与开源积累，形成少样本学习、强化学习与元学习方向的长期贡献。
Contact
联系方式
Signals, profiles, and collaboration channels.
信号、主页与合作渠道。
For research collaboration, speaking, media, or company-building conversations.
适用于研究合作、演讲、媒体与创业交流。