This research group develops tools that enable intelligent systems to
- extract, e.g., sampling[ 1 ], sensor selection[ 2 ], experimental design[ 3 ]
- process, e.g., constrained learning[ 4 , 5 , 6 , 7 ], network data processing[ 8 , 9 ], estimation/inference[ 10 , 11 , 12 , 13 ]
- act, e.g., resource allocation[ 14 ], scheduling[ 15 ], resilient control[ 16 , 17 ], (safe) reinforcement learning[ 18 , 19 ]
on information.
Currently, the main drive of my research is developing the theory, algorithms, and applications of constrained learning, a tool that enables the data-driven design of systems that satisfy requirements such as robustness[ 4 , 5 , 6 , 7 ], fairness[ 4 , 7 ], safety[ 18 , 19 , 20 ], smoothness[ 21 ], and invariance[ 22 ]. On the one hand, I investigate fundamental questions such as
- when is it possible to learning under requirements? (whenever you can learn at all)[ 4 , 7 ]
- how much harder is it than vanilla learning? (essentially the same difficulty)[ 4 , 7 ]
- are there problems that only constrained learning can tackle? (in short, yes)[ 20 ]
On the other hand, I am interested in the impact constrained learning can have on traditional learning tasks, such as image classification[ 4 , 5 , 6 , 7 , 22 ], semi-supervised learning[ 21 ], and data-driven control[ 18 , 19 , 20 ]. Most importantly, I think of constrained learning as a new mindset for the design data-driven solutions shifting away from the current objective-centric paradigm towards a constraint-driven one.
- Dr. Luiz Chamon