By Grace Stanley
Tianyi Chen is pushing the boundaries of artificial intelligence by asking a pressing question: What if AI could be engineered not just to optimize for a single outcome, but to make smarter, more balanced decisions — much like humans do?
Chen, a new associate professor of electrical and computer engineering at Cornell Tech and Cornell Engineering, leverages deep mathematical insights to design algorithms that help AI juggle multiple priorities at once — accuracy, fairness, efficiency, and reliability — rather than optimizing for just one.
This balanced approach could make generative AI tools more reliable, strengthen large-scale computing systems, and improve the energy efficiency of the new AI chips that power them. His work has already led to patented innovations through collaborations with IBM.
In addition, Chen is developing a new research direction he calls “Physical neural networks-for AI,” which explores how the laws of physics can be harnessed to perform energy-efficient computation. He sees New York City as a “living laboratory” for testing these ideas and forming industry partnerships.
Chen joins Cornell Tech from Rensselaer Polytechnic Institute (RPI), where his research was supported by the RPI–IBM Artificial Intelligence Research Partnership. He holds a bachelor’s degree from Fudan University and a Ph.D. from the University of Minnesota. His accolades include the Institute of Electrical and Electronics Engineers Signal Processing Society Best Ph.D. Dissertation Award, an NSF CAREER Award, and industry honors from Amazon and Cisco.
In the Q&A below, read more about Chen’s motivations, research, and what excites him about joining the Cornell Tech faculty.
What motivated you to come to Cornell Tech?
I am deeply motivated by Cornell Tech’s unique model, which merges the rigorous academic atmosphere of Cornell with the vibrant tech ecosystem of New York City. My research, which focuses on translating foundational theory of optimization, machine learning, and signal processing into real-world solutions, aligns perfectly with this spirit. My collaborations with IBM, for instance, have improved their speech models and are shaping the understanding of training AI models on new energy-efficient analog AI chips.
Cornell Tech’s emphasis on interdisciplinary collaboration and its deep integration with the New York City tech ecosystem provide the ideal environment to not only advance the theory and algorithms of multi-objective AI but also to see it deployed in practice to solve real challenges.
What are you most looking forward to about working in New York City?
I’m excited to use New York City as both a living laboratory and a global hub for innovation. My research on algorithmic advances for analog AI computing directly connects to the state’s growing role in semiconductor research and development. At the same time, the concentration of world-class tech and financial companies in Manhattan offers unparalleled opportunities to test my algorithms in real-world settings, form new industry partnerships, and ensure my work has both scientific and commercial impact.
What is your academic and research focus?
My current research focuses on developing principled algorithms for cross-layer designs of more reliable, robust, and efficient AI systems. Today’s AI models and algorithms were largely designed to optimize for a single metric, like accuracy, but achieving human-like intelligence requires balancing multiple competing objectives simultaneously.
To address this, I build new theories to understand the trade-offs among competing performance metrics and design provably efficient algorithms based on multi-objective and bilevel optimization. These advances have direct applications to generative AI post-training, large-scale distributed and federated learning, and next-generation AI hardware.
What inspired you to pursue a career in this field?
My path began with a childhood love of math and physics, but it was in university that I discovered the excitement of applying these tools to real engineering challenges. The field of electrical and computer engineering offered me a way to bridge elegant mathematical theory with tangible societal impact. Today, by collaborating with colleagues in different domains, my work spans from advancing AI safety and robustness to developing methods for accurate training and inference on imperfect analog AI chips.
What scientific questions are you looking to answer next?
I am exploring a new paradigm I call “Physical neural networks for AI,” where the laws of physics themselves are harnessed to perform computation and generate intelligence. A central question is how complex AI tasks — including multi-objective learning — can be mapped onto physical systems. My recent work on algorithms for analog neural networks is just the first step. Ultimately, I aim to create a feedback loop where AI not only runs on physical substrates but also accelerates their design.
Why Cornell?
I chose Cornell because of its legacy of academic excellence and its strong group of world-class faculty members and talented students. As a Cornell faculty member, I am excited to collaborate with world-class colleagues and mentor talented graduate students, both in the classroom and on research that pushes the boundaries of balanced, multi-objective AI. In addition, Cornell, and particularly the Cornell Tech campus, fosters a unique ecosystem where this kind of translational research is not just supported but is central to its mission.
Grace Stanley is the staff writer-editor for Cornell Tech.
