Introducing the 2026 CBI fellows

This year’s CBI fellowship applications provided a strong and wide-ranging pool of candidates for the postdoctoral program. We are proud to announce this year’s cohort of six accomplished researchers whose academic successes, professional passion, and leadership potential made them stand out to our review committee as exceptionally aligned with the fellowship’s goal: to provide emerging research talents with the resources to expand their professional horizons and make a lasting positive impact in their careers and, ultimately, in society.

Yihao Chen

Chen’s research focuses on building trustworthy AI systems for internet security and operations, particularly in real-world environments where observations are partial, adversarial, and operational constraints dominate. His research lies at the intersection of AI models, semantic representation, and anomaly detection in network security.

While at CMU, he intends to investigate the development of agentic systems to analyze security incidents at internet scale by taking advantage of deep (semantic) security understanding. He is also interested in the creation of trustworthy datasets that will support the training of dedicated and unbiased AI models optimized for security operations and network security tasks.

Chen’s research received the prestigious Internet Defense Prize and was recipient of the Distinguished Paper Award at the 33rd USENIX Security Symposium in 2024. Chen received B.S. and Ph.D. degrees in computer science from Tsinghua University in Beijing, China.

He will be hosted by Vyas Sekar, professor of electrical and computer engineering.

Eliahu Horwitz

Eliahu Horwitz’s research solves a major challenge in the world of AI: navigating and finding the right model among the millions already trained. Model developers spend time, energy, and money training new models because it is impractical to search through existing models. Instead, Eliahu's work treats neural networks–the fundamental building blocks of AI models–as data, and defines novel techniques to analyze, retrieve, and efficiently generate AI models.

While at CMU, he intends to work with a wide range of collaborators to build a "model atlas:" a map of the world-wide AI landscape that charts models, their attributes, and the model weight transformations that connect them. Further, Horwitz will define the machine learning tools necessary to effectively search the model atlas for model reuse opportunities. 

His research has been published in top-tier venues including CVPR, ICLR, ICML, ICCV, and NeurIPS and recognized with the Google PhD Fellowship in Machine Learning and ML Foundations. Eliahu earned his M.S. and Ph.D. degrees in computer science from The Hebrew University of Jerusalem under the supervision of Associate Professor of Computer Science and Engineering Yedid Hoshen.

He will be hosted in the Computer Science Department by Jun-Yan Zhu, associate professor of computer science and robotics.

Romina Mir

Romina Mir is a roboticist whose research bridges machine learning, dexterous manipulation, and bio-inspired motor learning. She earned her Ph.D. in biomedical engineering from the University of Southern California, where her research drew inspiration from biology to investigate concepts such as curriculum-based learning, sensory feedback, and mechanical compliance into robotic manipulation. Her work examined how these principles can enable more effective learning in contact-rich robotic tasks.

During her tenure at CMU, Mir will develop force-aware retargeting and representation learning methods for contact-rich dexterous manipulation, with a particular focus on cross-embodiment learning and skill transfer. Her work seeks to enable robots to transfer manipulation skills more reliably across different robotic hands and embodiments, objects, and real-world settings, ultimately advancing robust and adaptable manipulation capabilities for industrial automation.

Her research has appeared in Science Advances and at the IEEE Humanoid Robots, where she was the top recipient of the Kanako Miura Award for Women in Engineering.

Mir will be hosted by Jeffrey Ichnowski, a professor in the Robotics Institute at CMU.

Maithili Patel

Maithili Patel is a Ph.D. candidate in robotics at the Georgia Institute of Technology whose research lies at the intersection of artificial intelligence and human-robot interaction. Her work develops methods that enable robots to anticipate users’ needs and personalize assistance from observations and sparse feedback, with publications at CoRL, IEEE Robotics and Automation Letters, and COLM.

At CMU, Patel will investigate proactive, longitudinal robot assistance: systems that adapt as users’ routines, preferences, and trust evolve over time. Her project will also develop context-aware methods that help robots choose and adapt actions in socially appropriate, transparent, and minimally disruptive ways, supporting trustworthy assistance in homes, healthcare settings, and other everyday environments.

Maithili received honorable mention for the Richard and Eleanor Towner Prize for Outstanding Graduate Student Instructors in 2020, and she received the Georgia Robotics Fellowship in 2024.

She will be hosted in the Language Technologies Institute by Yonatan Bisk, assistant professor in the Language Technology Institute.

Aafaq Sabir

Aafaq Sabir’s work lies at the crossroads of privacy, transparency, and compliance enforcement in emerging AI-driven smart home ecosystems and voice assistants. His research examines how these systems interact with user expectations and behaviors, with a focus on understanding the gap between user mental models and actual system behavior. The final goal of this work is to closing the gap between the user’s understanding and the system’s behavior, and making the smart voice-assistant ecosystems trustworthy and policy-compliant.

While at CMU, he intends to continue to investigate transparency, trust, and human agency challenges in emerging AI and smart homes ecosystems, and to extend his studies to industrial and automotive environments. He will extend his work to the identification of novel attacks and the design of novel LLM safety protections that can be applied to multi-modal smart ecosystems.

Sabir’s research has been published at top security, privacy, and usability venues, including Usenix security, PETS, IEEE Security and Privacy Symposium, and ACM CHI. His  research has received news and media coverage in outlets such as WRAL, Social-MediaToday, and Customer Data Platform Institute, and also received the Best Paper Runner-up award at PETS 2026. Sabir received M.S. and Ph.D. degrees in computer science from North Carolina State University (NCSU) under the supervision of Anupam Das, assistant professor of computer science.

Sabir will be hosted by Lujo Bauer, professor of electrical and computer engineering.

Thomas Zhang

Thomas Zhang is a researcher at the intersection of machine learning, control, and optimization, completing his Ph.D. in electrical and systems engineering at the University of Pennsylvania. His work develops theoretical and practical foundations for reliable learning in dynamical systems, with publications at ICLR, ICML, and NeurIPS.

At CMU, Zhang will study how to make learning-based agents (including robots and LLM agents) more robust under long-horizon feedback and distribution shift. His project will develop optimization methods, model designs, and evaluation criteria that prioritize real-world, closed-loop performance rather than training loss alone, advancing more predictable and adaptable autonomous systems.

Zhang’s research was recognized as a Spotlight presentation at the prestigious ICLR conference in 2024, and he was awarded the AWS-AI ASSET Fellowship in 2025.

Zhang will be hosted in the Machine Learning Department by assistant professors Aditi Raghunathan and Max Simchowitz.