The nexus of learning, optimization, and the leading edge of practice for chip design, networks, and robotics.

TILOS Industry Day on June 2, 2025 at UC San Diego's Halıcıoğlu Data Science Institute welcomed industry speakers and panelists from Intel, Brain Corp, Agility Robotics, Microsoft Research, Amazon Web Services, NVIDIA, Qualcomm, IBM, and Calibrate Ventures to offer their perspectives on challenges in AI, optimization, and the use domains of chip design, networks, and robotics. The packed agenda also featured research posters and lightning talks from TILOS graduate students and postdoctoral scholars, TILOS faculty research highlight talks, and two panel discussions: AI Challenges for Academia—An Industry Perspective and Building Deep Tech Companies.

TILOS welcomed two new postdoctoral scholars in July 2025. Dr. Zhiyang Wang joins UC San Diego after earning her PhD from the University of Pennsylvania, where she was advised by TILOS Networks team co-lead Alejandro Ribeiro. Dr. Xiaohan Gao is new to TILOS and joins the University of Texas at Austin, where she works with TILOS Chips team co-lead David Pan, as well as Director Yusu Wang.

Dr. Zhiyang Wang

UC San Diego

Where did you earn your graduate degree?

I got my PhD in the Department of Electrical and Systems Engineering at the University of Pennsylvania in May 2025, advised by Prof. Alejandro Ribeiro.

What are your research interests?

My research interests encompass graph signal processing, graph neural networks, geometric data analysis, wireless communications, and scalable autonomous systems. My research builds the mathematical foundations to understand the fundamental properties of graph neural networks via a manifold perspective. These theoretical understandings can advance scalable and stable solutions in many graph-structured applications such as communication networks, robotic systems, and data manifolds. My ultimate goal is to use insights from signal processing to characterize mathematical foundations, develop novel architectures, and devise alternative training procedures for deep learning over real-world geometric data.

Which faculty member(s) are you working with as a postdoc?

I work with Prof. Yusu Wang on the generalization of learning models for graphs and sets, focusing both on methodology and on the application front.

What type of career will you pursue after your current position?

I will join the Electrical and Systems Engineering Department at Washington University in St. Louis as an Assistant Professor in Fall 2026.

Any recent publications or awards to highlight?

We recently had a paper accepted to ICML 2025 on the generalization of graph convolutional neural networks across graph scales. We introduce a manifold-based view that unifies discrete graphs and continuous geometry. Spectral insights reveal how to balance discriminability vs. generalization as GNNs scale. 

What do you enjoy doing when not working on research?

I enjoy hitting the gym, exploring trails, building LEGO, and diving into Genshin Impact.

Dr. Xiaohan Gao

UT Austin

Where did you earn your graduate degree?

I obtained my PhD degree from Peking University, and I conducted research on electronic design automation (EDA).

What are your research interests?

I have broad research interests in EDA and optimization algorithms. I am particularly interested in the application of machine learning to EDA problems. My past research mainly defines problems in EDA as optimization problems and develop corresponding optimization algorithms to solve them. In addition, I am personally very interested in high-performance computing and programming languages.

Which faculty member(s) are you working with as a postdoc?

I work with Professors David Pan and Yusu Wang.

What type of career will you pursue after your current position?

I'm still looking for future opportunities, and I tend to pursue an academic position.

Any recent publications or awards to highlight?

I won the Best Paper Award at 2023 Design, Automation and Test in Europe Conference.
In that paper, we propose a routing framework considering electronic constraints for analog layout automation. Also, I won best poster award of ACM SIGDA Student Research Forum at ASP-DAC 2025 and shared about our framework on analog circuit design automation. Our framework consists of both front-end schematic design and back-end layout design with consideration of designer interaction.

What do you enjoy doing when not working on research?

I enjoy playing baseball. I'm also a huge sports fan, and I love watching various kinds of sporting events, including F1, football, tennis, esports, etc.

Recordings of all TILOS Seminars and other presentations are at tilos.ai/recorded-talks and the TILOS YouTube channel.

New algorithms enable efficient machine learning with symmetric data

A new study led TILOS Foundations team member Stefanie Jegelka and graduate student Behroozi Tahmasebi presents the first provably efficient method for machine learning with symmetric data, which could advance neural network design for applications ranging from drug discovery to climate modeling. Full story in MIT News.

Unpacking the bias of large language models

A team of MIT researchers, including TILOS Foundations team member and associate professor Stefanie Jegelka, and postdoc Yifei Wang, has developed a theoretical framework to study how information flows through the machine learning (ML) architecture that forms the backbone of LLMs. Their work has uncovered the root cause of "position bias" in LLMs, paving the way for more accurate and reliable AI systems. Full story in MIT News.

Congratulations to UCSD graduate student Kehan Long for being accepted to the 2025 RSS Pioneers Workshop!

Congressional Robotics Caucus relaunches to help U.S. industry

The Congressional Robotics Caucus had not convened formally since 2019, according to TILOS Robotics team co-lead Henrik Christensen, who is also the Qualcomm Chancellor’s Chair of Robot Systems and a distinguished professor of computer science at the Department of Computer Science and Engineering at UC San Diego. Full story in The Robot Report.

Opinion: We Can’t Regulate Our Way to Crypto Leadership. We Still Need Science

Steering AI: New Technique Offers More Control Over LLMs

Imagine developing a finer control knob for artificial intelligence (AI) applications like Google Gemini and OpenAI ChatGPT. Mikhail Belkin, a professor with UC San Diego’s Halıcıoğlu Data Science Institute (HDSI) has been working with a team that has done just that. Specifically, the researchers have discovered a method that allows for more precise steering and modification of large language models (LLMs). Belkin said that this breakthrough could lead to safer, more reliable and more adaptable AI. Full story in UC San Diego Today.

MORE NEWS >

TILOS research publications are available on Google Scholar.

Stay connected with TILOS