Carnegie Mellon University
Machine Learning Department - AI Safety & Robustness Research
United States · PhD/Postdoc
The Machine Learning Department at Carnegie Mellon University is the world's first independent ML department, with deep expertise in AI safety and robustness research. Research directions encompass adversarial robustness, out-of-distribution generalization, trustworthy ML systems, and safe reinforcement learning, providing a complete research chain from theoretical analysis to system implementation for AI alignment.
学术历史
CMU established the world's first Machine Learning Department in 2006, founded by Tom Mitchell. AI safety research at CMU has a long history: Matt Fredrikson's adversarial example research, Pradeep Ravikumar's robust ML theory, and David Held's robot safety constitute a multi-dimensional safety research matrix. CMU holds a unique position: it combines the strong technical foundations of the ML Department with the security engineering tradition of the Software Engineering Institute (SEI), giving its AI safety research both theoretical depth and engineering practicality.
当前状态
The CMU ML Department currently has multiple research groups involved in AI safety: adversarial robustness and defense, out-of-distribution generalization, trustworthy AI, and safe reinforcement learning. It hosts an AI Safety Reading Group and regular seminars. CMU also participates in the NIST AI Safety Institute Consortium, providing technical support for United States AI safety policy. Courses such as 10-708 (Probabilistic Graphical Models) cover safety and uncertainty topics.
实验室 / 研究中心
Machine Learning Department
访问页面 →关键人物
Matt Fredrikson
Associate Professor
Expert in AI safety and privacy research, focusing on adversarial examples, model inversion attacks, and defenses. His adversarial example research is among the most highly cited work in the field.
Pradeep Ravikumar
Associate Professor
Expert in robust machine learning theory, researching distributionally robust optimization, interpretable ML, and statistical learning theory, providing theoretical guarantees for safe ML.
David Held
Assistant Professor
Expert in robot safety and manipulation, researching the application of safe reinforcement learning in robotics and safety constraints in human-robot collaboration.
标志性成果
- Systematic research on adversarial examples: the C&W attack method has become a standard benchmark for adversarial robustness evaluation
- Theoretical framework for distributionally robust optimization (DRO), providing formal guarantees for ML system safety under distribution shift
- Model inversion attack research, revealing privacy security risks in ML models and advancing privacy protection technologies
- Application of safe reinforcement learning in robotic manipulation: constraint satisfaction and safe exploration algorithms
学术资源
证据
师资
7 位相关教师
知名：Matt Fredrikson、Pradeep Ravikumar、David Held、Ruslan Salakhutdinov、Louis-Philippe Morency
研究产出
CMU ML Department researchers consistently publish AI safety and robustness papers at top conferences and security venues including NeurIPS, ICML, ICLR, and S&P
就业去向
Graduates join companies such as Google DeepMind, OpenAI, Anthropic, and NVIDIA, or take faculty positions at leading universities
发现信息有误？ 提交纠错