Penn ESE Academic Job Market Candidates
Our department is proud to introduce members of our community who are searching for faculty positions.
Please contact Elizabeth Kopeczky for reference requests.
Luiz F. O. Chamon
Bio: Luiz Chamon received the B.Sc. and M.Sc. degrees in electrical engineering from the University of São Paulo, São Paulo, Brazil, in 2011 and 2015 and the Ph.D. degree in electrical and systems engineering from the University of Pennsylvania (Penn), Philadelphia, in 2020. He is currently a postdoc of the Department of Electrical and Systems Engineering of the University of Pennsylvania. In 2009, he was an undergraduate exchange student at the Masters in Acoustics of the École Centrale de Lyon, Lyon, France. In 2009, he was an Assistant Instructor and Consultant on nondestructive testing at INSACAST Formation Continue. From 2010 to 2014, he worked as a Signal Processing and Statistical Consultant on a project with EMBRAER. In 2018, he was recognized by the IEEE Signal Processing Society for his distinguished work for the editorial board of the IEEE Transactions on Signal Processing. He also received both the best student paper and the best paper award at IEEE ICASSP 2020. His research interests include optimization, signal processing, machine learning, statistics, and control. More information available at http://www.seas.upenn.edu/~
Learning under requirements
The transformative power of learning lies in automating parts of the engineering of complex systems, allowing us to go from data to operation with little to no human intervention. These data-driven solutions, however, often lead to biased, prejudiced systems prone to tampering (adversarial examples) and unsafe actions. To make this autonomous engineering vision a reality, we must advance beyond the current paradigm of minimizing costs to develop methods capable of learning under requirements. As in classical learning, this problem can be cast as an optimization program, albeit one with constraints. Yet, this distinction is a major source of complexity even in classical optimization. For instance, whereas linear regression is trivial, constraining the coefficients to be sparse makes the problem NP-hard. Our current learning theory is not equipped to address the questions of when and how it is possible to learn under requirements.
In this talk, I will answer these questions by developing the theoretical underpinnings of constrained learning. I will introduce a constrained learning framework that extends the classical probably approximately correct (PAC) one and show that despite appearances, constrained learning is not harder than unconstrained learning, i.e., they have essentially the same sample complexity. Additionally, I will present a practical learning rule that under mild conditions can tackle constrained learning tasks by solving only unconstrained empirical risk minimization (ERM) problems. I will demonstrate how these advances address problems in fair classification, robust image recognition, and safe reinforcement learning. I see these contributions as a first step towards a shift from the current, objective-centric paradigm to a constraint-driven learning one that I will briefly discuss together with the new theoretical and practical questions it raises.
Bio: Fernando Gama is a Postdoctoral Scholar with the Electrical Engineering and Computer Sciences department at the University of California, Berkeley. He received a Ph.D. in Electrical and Systems Engineering from the University of Pennsylvania in 2020, a M. A. in Statistics from the Wharton School in 2017, and an Electronic Engineering degree from the School of Engineering of the University of Buenos Aires, Argentina in 2013. He was a visiting researcher at TU Delft in 2017, and a research intern at Facebook Artificial Intelligence Research in 2018. He was awarded a Fulbright scholarship for international students for 2014-2016. His research interests are in the field of information processing and machine learning over network data. More information available at http://gama.ar.
Graph Neural Networks and Collaborative Intelligent Systems
Graphs are generic models of signal structure that can help to learn in several practical problems. To learn from graph data, we need scalable architectures that can be trained on moderate dataset sizes and that can be implemented distributedly. In this talk, I will draw from graph signal processing to define graph convolutions, and use them to introduce graph neural networks (GNNs). I will prove that GNNs are permutation equivariant and stable to perturbations of the graph, properties that guarantee scalability and transferability. I will also use these results to explain the advantages of GNNs over linear graph filters. I will then discuss the problem of learning decentralized controllers, and how GNNs naturally leverage the partial information structure inherent to distributed systems. Using flocking as an illustrative example, I will show that GNNs, not only successfully learn distributed actions that coordinate the team, but also transfer and scale to larger teams.
Bio: Yiannis Kantaros received the Diploma in Electrical and Computer Engineering in 2012 from the University of Patras, Patras, Greece. He also received the M.Sc. and the Ph.D. degrees in Mechanical Engineering from Duke University, Durham, NC, in 2017 and 2018, respectively. He is currently a postdoctoral researcher h in the Department of Computer and Information Science at the University of Pennsylvania. His research focuses on distributed control, machine learning, and formal methods with applications to distributed robotics. He received the Best Student Paper Award at the 2nd IEEE Global Conference on Signal and Information Processing in 2014 and the 2017-18 Outstanding Dissertation Research Award from the Department of Mechanical Engineering and Materials Science, Duke University, Durham, NC.
Safe & Distributed Autonomy in Unknown Environments
Recent advances in machine learning, computer vision, and control theory offer a tremendous opportunity to deploy autonomous robot systems to uncharted environments to accomplish complex missions. Such tasks are particularly challenging as they require the robots to operate in environments with unknown structure, degraded environmental conditions, severe communication and sensing constraints, and expansive areas of operation. With the goal of ultimately designing safe and robust autonomous robotic systems that can operate in these conditions, I will first present a novel perception-based control framework that allows multi-robot systems to accomplish complex missions in unknown environments with safety guarantees. The proposed approach generates reactive control policies that adapt to the continuously learned map of the environment that is updated using learning-based perception systems. To enhance robustness of the learning-based perception component against adversarial attacks, I will also present a new attack-agnostic defense mechanism that can detect in real time if the input images to the perception system have been manipulated adversarially by digital (noise at the pixel level) or physical attacks (adversarial stickers in the physical world). Finally, I will also present a novel distributed autonomy framework that enables robot teams that reside in environments with severe communication constraints to accomplish complex collaborative mission and safety requirements.