Yale awarded 18 AI Seed grants through their AI at Yale initiative. the 18 teams comprised of 62 faculty and research personnel from across FAS, SEAS, YSE, YSL, YSA, and YSM receiving this year’s awards. The projects include significant graduate student and postgraduate involvement.
The department of physics, through primary and secondary faculty members, is involved with four of the awards. Learn about their projects below.
The names of Principal Investigators are in bold. Those affiliated with physics are starred.
Ensemble Stacked Deep Learning networks for Prognonsis of the Deadliest Brain Cancer
*Andre Levchenko, Professor Biomedical Engineering and Physics
Murat Gunel, Professor Neurosurgery, Neuroscience and Genetics; Chair Neurosurgery
Karthik Desingu, Graduate Student, Biomedical Engineering
Project: Predicting cancer outcomes is crucial, especially for aggressive cancers like glioblastoma, which has a median survival of 1.5 years. Current tests are imprecise. Our project will use AI to create a precise, patient-specific prognostic test by integrating diverse clinical, genetic, and phenotypic data, improving predictions and patient care quality.
Levchenko explains the project, “Glioblastoma is a particularly common and aggressive brain cancer. It is not curable and, on average, patients only survive for a year and a half following diagnosis. However, there are also long-term survivors, who may live with the disease for several years. At present, we do not have a prognostic test that may predict the expected survival for a specific patient in terms of months and years, which substantially worsens both clinical options and the quality of life. The proposal from Levchenko lab is based on the prior and current extensive collaborative work with the Yale Neurosurgery department. It resulted in extensive datasets experimentally obtained from a large patient cohort, both based on new tests at the Levchenko lab and as a part of the clinical process. In particular, the Levchenko lab has shown that the way the cancer cells obtained from a patient tumor move can be predictive of how the patient will survive. This very surprising finding suggests that the biophysical properties of the cells underlying cell migration determine the eventual clinical outcomes in this cancer.
Although many of the clinical or lab data can be partially predictive of the clinical outcomes, such as the overall survival, none of the tests is fully predictive on its own. Previously, several attempts have been made to build AI-based models to use some clinical data to improve the predictive power, but again, with limited success. In their proposal, Levchenko and his clinical collaborators lay out a new approach that will create a super-model, combining or stacking the existing models and datasets into one hybrid whole. The advantage of such stacked neural network models is that they integrate all the data available, in contrast to the models that would only be based on the analysis e.g., of clinical images or of the tissue sections. The stacked models are more universal and flexible, which has already been shown to lead to much better prediction and classification in the clinical analysis of other diseases. Levchenko hopes that this will be the case with glioblastoma, based both on their novel biophysical and bioengineering lab tests and more mainstream biological and clinical data. If successful, this project can lay the foundation for a fundamental improvement of how we analyze and, ultimately, treat this and other aggressive and currently incurable cancers.”
A Geometric Solution for Optimally Quantum-Efficient Photosynthesis Evolved in Giant Clams
Alex Wong, Assistant Professor, Computer Science
*Alison Sweeney, Associate Professor, Ecology and Evolutionary Biology and Physics
Project: We will develop a solution for land-, water-, and carbon-efficient algal biomass production using sunlight, inspired by giant clams. The Sweeney group found these clams optimize sunlight conversion through dynamically organized microalgae arrays. Using AI-assisted computer vision, we will study the clams’ adaptations to light changes. This knowledge will help create highly efficient engineered materials for biomass production, potentially reducing land use significantly. Our goal is to track the microalgae positions with high precision, mimicking the clams’ efficiency under varying light conditions.
Interpretable AI for Predictive Spectroscopy of Correlated Quantum Materials
Tianyu Zhu, Assistant Professor, Chemistry
*Yu He, Assistant Professor, Applied Physics and Physics
Ke Liao, Postdoctoral Associate, Chemistry
Christian Venturella, Graduate Student, Chemisty
Jinming Yang, Graduate Student, Applied Physics
Project: We aim to develop an AI framework to predict the spectroscopic and superconducting properties of complex quantum materials, focusing on copper oxide superconductors (cuprates). Traditional physics methods struggle with these materials’ strong electron interactions. Our project integrates advanced spectroscopy, quantum simulations, and deep learning. Using a physics-informed model, we will predict electron spectra to uncover superconducting phases. We will also create a unique dataset for AI training, aiding in discovering new superconductors and advancing materials research.
Yu He comments about the project, “High-temperature superconductors are remarkable materials that can conduct electricity with zero resistance at relatively high temperatures, offering exciting possibilities for energy-efficient electronics, powerful magnets, and controlled plasma fusion. Yet, understanding and predicting their behavior remains one of the grand challenges in science. These materials are governed by extremely strong interactions between electrons, making their collective behavior too complex for traditional theories or standard computational models to capture. As a result, designing new superconductors with improved performance has largely relied on trial and error rather than predictive insight.
This project brings together two complementary approaches: advanced experiments and powerful computational methods, integrated through a new form of artificial intelligence (AI). On the computational side, Professor Tianyu Zhu’s group will develop cutting-edge simulations of electron interactions using state-of-the-art quantum many-body techniques accelerated by machine learning. On the experimental side, Prof. Yu He’s group will use a technique called angle-resolved photoemission spectroscopy (ARPES), which measures the energy and motion of electrons in exquisite detail. These experimental data - helped by AI classifiers - will generate a rich, multi-dimensional picture of how electrons behave in strongly correlated materials. By combining simulations and experiments, Zhu’s team will build interpretable, physics-informed ML models that can predict how electrons respond to changes in material composition and structure, effectively connecting theory with experimental data. The ultimate goal is to create a workflow where experimental data and computational predictions continuously inform each other, enabling the discovery of new superconductors in a guided, efficient way. Beyond identifying promising materials with higher superconducting transition temperatures, the project will produce an open-access dataset and AI tools for the broader scientific community. In this way, the team hopes to establish a new, interpretable AI-driven framework for understanding and designing complex quantum materials, moving the field from observational science to predictive discovery.”
Physics-informed AI for eigenvalue problems and astrophysical applications
Earl Bellinger, Assistant Professor, Astronomy
*Daisuke Nagai, Professor, Physics and Astronomy
Lu Lu, Assistant Professor, Statistics and Data Science
Sifan Wang, Postdoctoral Associate, Statistics and Data Science
*Naomi Gluck, Graduate Student, Physics
Project: We explore how artificial intelligence, especially deep learning, can improve the solving of eigenvalue problems, which are essential in fields like quantum mechanics, structural engineering, and stellar astrophysics. By overcoming current AI limitations, we aim to boost scientific research across many areas. This advancement will help us use physics-informed neural operators to solve these problems more accurately and comprehensively, speeding up discoveries and innovations in science.
An interdisciplinary team of researchers from the departments of Physics, Astronomy, and Statistics and Data Science (S&DS) has secured a 2025 Yale Seed Grant for their pioneering research aimed at bridging a critical gap between artificial intelligence and fundamental scientific computation. The team—comprised of Professor Earl Bellinger (Astronomy), Professor Daisuke Nagai (Physics), Professor Lu Lu (S&DS), Dr. Sifan Wang (S&DS), and graduate student Naomi Gluck (Physics) —is developing a breakthrough class of AI solvers to tackle one of science’s most ubiquitous challenges: the eigenvalue problem.
The Universal Challenge of Eigenvalue Problems
Eigenvalue problems are foundational to modern science and engineering, acting as the mathematical “fingerprints” of physical systems. Their solutions describe critical properties across immense scales, from the energy levels of quantum particles to the resonant vibrations of stars, planets, and bridges. Despite their importance, current artificial intelligence techniques, including supervised models and Physics-Informed Neural Networks (PINNs), critically lack the precision demanded by scientific applications.
A Novel Framework: Physics-Informed Deep Neural Operators
The team’s research introduces an innovative solution: semi-supervised Physics-Informed Deep Neural Operators (DNOs). This next-generation machine learning architecture represents a paradigm shift. It uniquely synergizes the pattern-recognition power of deep learning with the rigor of physical laws and the hard constraints of explicit eigenvalue data. This hybrid approach is designed to achieve a significant leap in both the accuracy and robustness required to solve complex physical problems.
“Our goal is to fundamentally elevate the capability of AI in the physical sciences,” explains Professor Nagai. “We are building tools that don’t just approximate solutions but compute them with the precision that scientific discovery demands.”
From the Heartbeat of Stars to Broad Scientific Impact
To validate and refine these new methods, the team is turning to the stars. Asteroseismology, the study of stellar pulsations, provides an ideal proving ground due to its stringent accuracy requirements and the wealth of high-precision observational data from missions like NASA’s Kepler spacecraft. With this seed funding, the project will enhance the accuracy of AI-driven solutions for stellar pulsations by two orders of magnitude—from the current 1% margin of error down to an exceptional 0.01%. This advancement will pave the way toward the 0.0001% precision needed to fully leverage data from modern space telescopes and unlock new discoveries about stellar evolution.
While its first application is in astrophysics, the project’s impact is expected to be far-reaching. The development of general-purpose, high-precision AI eigenvalue solvers promises to accelerate innovation across a multitude of fields, including quantum mechanics, materials science, structural engineering, and fluid dynamics. This grant provides vital support for graduate student Naomi Gluck and establishes the foundational work necessary to pursue larger-scale federal funding, positioning Yale at the forefront of AI-driven scientific discovery.