Team & Tasks

This page outlines the project team, thrust leadership, key collaborators, and the specific tasks defined under each research thrust. The structure was designed to foster interdisciplinary collaboration across machine learning, robotics, verification, and developmental psychology, drawing inspiration from child learning principles.

Project Leadership & Key Personnel

Principal Investigator (PI)

• Insup Lee (University of Pennsylvania): Cyber-Physical Systems (CPS), High-Assurance Machine Learning, Security

Co-Principal Investigators & Thrust Leads

• Osbert Bastani (Thrust I Lead): Machine Learning, AI Safety, Programming Languages
• Eric Eaton (Thrust II Lead): Machine Learning, Lifelong/Continual Learning, Interactive AI
• James Weimer (Thrust III Lead): Learning-Enabled CPS, Autonomous Systems, Verification
• Insup Lee (Thrust IV Lead): Integration, Evaluation, CPS Assurance

Key Collaborators

• Kostas Daniilidis: Computer Vision, Robotics, Machine Learning
• Dan Roth: Machine Learning & Inference, Natural Language Processing
• Julia Parish-Morris: Developmental Psychology, Language Development, Childhood Learning

Additional Team Members (at various points during the project)

• PhD Students: Meghna Gummadi, Sooyong Jang, Ramneet Kaur, Shuo Li, Vivian Lin, Stefanos Pertigkiozoglou, Kaustubh Sridhar, Yahan Yang, Matthew Cleaveland
• Postdoctoral Researchers: Michele Caprio, Souradeep Dutta, Kuk Jin Jang, Georgios Georgakis
• Research Assistants: Martin Ricardo Del Rio Grageda, Maxine Covello
• Former Members: Soham Dan (now at IBM), Radoslav Ivanov (RPI), Yiannis Kantaros (WUSTL), Ivan Ruchkin (UF)

For the most up-to-date individual contributions and publications, please visit the respective faculty lab pages or academic profiles.

Thrusts and Tasks

The project was organized into four thrusts, each with defined tasks and connections to principles of child learning.

Thrust I: Concept-based Learning Robust to Adversarial Examples

• Lead: Osbert Bastani
• Personnel/Collaborators: Osbert Bastani, Kostas Daniilidis, Edgar Dobriban, Eric Eaton, Insup Lee, Julia Parish-Morris, Dan Roth, James Weimer
• Task I.1: Robust learning of visual object and scene representations
• Task I.2: Concept-based deep learning with inductive biases
• Task I.3: Compositional inference and reasoning for adversarial learning
• Connection to child learning: Concept selection and representation

Thrust II: Adaptive Learning in Dynamic Environments

• Lead: Eric Eaton
• Personnel/Collaborators: Osbert Bastani, Kostas Daniilidis, Julia Parish-Morris, Dan Roth
• Task II.1: Leveraging Inductive Biases for Adaptive Concept Learning
• Task II.2: Lifelong Learning
• Connection to child learning: Hierarchical and continual learning in new settings

Thrust III: Verification and Monitoring of Learning

• Lead: James Weimer
• Personnel/Collaborators: Insup Lee, Julia Parish-Morris, Oleg Sokolsky, James Weimer
• Task III.1: Verified learning
• Task III.2: Monitoring learning
• Connection to child learning: Trust building, validation by probing, self-adversarial techniques for robustness

Thrust IV: Integration and Evaluation

• Lead: Insup Lee
• Task IV.1: Toolset and dataset development
• Task IV.2: Evaluation platform and scenarios (including robotic testbeds and battlefield surrogate setups)

This structure supported cross-thrust integration, with evaluations drawing on real-world robotic platforms and demonstrations (e.g., ICRA 2022, CoRL 2022). See other tabs for detailed accomplishments, results, and impacts from these efforts.