AI-Powered Materials Seminar Series
As a follow-up to the AI-Powered Materials Discovery, Design and Synthesis Ideas Lab event, ICDS is hosting a three-part AI-Powered Materials Seminar Series.
Penn State faculty members will present materials research problems that could benefit from collaborations with artificial intelligence (AI) and machine learning (ML) experts. The event target audience includes faculty, staff and students with AI/ML expertise who are interested in starting new partnerships and collaborations with materials researchers on the presented challenges. The goal is to pair every speaker with at least one collaborator by the end of the series.
Each lecture will take place from 11 a.m. to 1 p.m. in Pollock Dining Commons Room 204, Penn State University Park. Lunch will be provided. Talks will start promptly at 11:30 a.m.; please arrive early to grab lunch.
Registration is encouraged, but not required. Feel free to attend even if not registered via the link below.
Seminar Schedule:
| Time | Agenda |
|---|---|
| 11 to 11:30 a.m. | Grab lunch from buffet and find a seat |
| 11:30 to 11:35 a.m. | Welcome from Vasant Honavar, ICDS Director of Strategic Initiatives; Professor of Informatics and Intelligent Systems |
| 11:35 to 11:55 a.m. | First Topic: Materials Researcher Presentation |
| 11:55 a.m. to 12:15 p.m. | First Topic: Discussion and Planning for Collaboration |
| 12:15 to 12:35 p.m. | Second Topic: Materials Researcher Presentation |
| 12:35 to 12:55 p.m. | Second Topic: Discussion and Planning for Collaboration |
| 12:55 to 1 p.m. | Closing from Vasant Honavar |
Dates and Speakers:
Wednesday, April 22, 2026
Christos Argyropoulos, associate professor in the Department of Electrical Engineering, presenting “AI-Powered Topology Optimization of Quantum Metamaterials”
ABSTRACT: Quantum nanophotonic metamaterials are engineered, artificial nanostructures made of periodic nanoscale unit cells that incorporate quantum material elements —such as entangled single photon pair emitting crystals and quantum dots. They are used to control and boost quantum light-matter interactions and manipulate quantum electromagnetic radiation at sub-wavelength scales. However, the design of such quantum metamaterials is extremely difficult due to their nanoscale features complexity and other fundamental physical limitations in the materials used, including the diffraction limit of light, poor efficiency due to extremely weak light-matter interactions, and decoherence/dephasing caused by radiative and nonradiative losses. In my talk, I will discuss about these problems and suggest ways to tackle them by using advanced AI-powered optimization algorithms combined with quantum electrodynamic simulations, with the goal to fabricate and realize new quantum metamaterials that use photons to efficiently encode and process quantum information. The primary goal will be to address key challenges like photon loss and decoherence, thereby enabling more robust and scalable quantum computing, communication, and sensing. To achieve this goal, we need to embark in a campaign of AI-assisted topology optimization combined with full-wave quantum optical simulations with the objective to fabricate and test new quantum metamaterial designs with improved functionalities.
Reginald Hamilton, Associate Professor of Engineering Science and Mechanics, presenting “Learning Thermoelastic Character: A Multimodal Processing-Structure-Property-Performance Framework for Shape Memory Alloy Design”
ABSTRACT: Shape memory alloys (SMAs) are a class of smart materials capable of reversible martensitic phase transformations that enable the shape memory effect and superelasticity. The global SMA market is currently estimated at $15–20 billion and is projected to grow substantially over the coming decades, driven by applications in biomedical devices, aerospace and defense systems, automotive and electric vehicles, soft robotics, smart infrastructure, consumer electronics, and emerging energy technologies. Despite this potential, SMA design remains challenging due to highly nonlinear thermomechanical behavior and strong sensitivity to composition and processing, making traditional trial-and-error approaches inefficient. The defining feature of SMAs is their thermoelastic transformation which cannot be reduced to a single property, but instead emerges from coupled phenomena including hysteresis, reversibility, energy dissipation, and transformation pathway stability. However, current machine learning approaches in SMAs largely focus on predicting transformation temperatures from composition, which does not capture the functional behavior required for design. This work introduces a new paradigm in which thermoelastic character is treated as a learnable, high-dimensional descriptor derived from multimodal experimental data. A processing–structure–property–performance framework is advanced in which thermoelastic character serves as the central design target. Multimodal data integrating processing conditions, microstructural features, bulk thermomechanical response, and full-field deformation measurements are used to establish linkages between structure and functional behavior. Full-field measurements provide spatially resolved mechanical signatures of transformation evolution, enabling interpretation of mesoscale heterogeneity not accessible through bulk metrics. By shifting from scalar property prediction to learning transformation behavior, this framework identifies processing–structure pathways associated with robust thermoelastic performance and provides a foundation for data-driven SMA design.
Tuesday, April 28, 2026
Linxiao Zhu, assistant professor in the Department of Mechanical Engineering, presenting “Opportunities in AI-driven design of photonic materials for energy and sensing”
ABSTRACT: Controlling light-material interaction using photonic materials is important for energy and information applications. In this seminar, I will talk about our research on radiative cooling, thermophotovoltaic thermal energy harvesting, and nonreciprocal photonics, and discuss opportunities in AI-driven design of photonic materials for these applications. First, I will talk about spectrally-selective design for passive radiative cooling, and for efficient thermophotovoltaic thermal energy harvesting. AI-driven inverse-design of photonic materials is critical to maximizing the performances. Second, I will talk about research on nonreciprocal photonic materials which points to dramatically new paradigm for sensing and energy conversion. To overcome limitations of existing nonreciprocal photonic materials, AI-powered design of quantum materials is greatly needed to accelerate the design of nonreciprocal photonic materials by combining AI and ab-initio modeling.
Darren Pagan, assistant professor of materials science and engineering and of mechanical engineering, presenting “AI/ML Opportunity in Materials: Identifying Rare Events to Predict Material Properties and Failure”
ABSTRACT: Many properties of materials are governed by rare events. For example, in polycrystalline metals, fatigue and fracture are often initiated in a single crystal (grain) which then propagates across the material. The traditional approaches for identifying the root causes of failure are primarily forensic, looking at the structure of the material after failure. However, new characterization techniques enable simultaneous tracking of the evolution of thousands of grains in situ (under applications of temperature, stress, electric field, etc.), possibly allowing for direct identification of the precursors to failure before they occur. The challenge lies in identifying rare events across large numbers of objects and feature sets with the presence of experimental uncertainty. Techniques are needed to effectively mine these new datasets to best predict material properties and constitute relationships.
Friday, May 1, 2026
Stephanie Law, Quantum Hub Associate Director, Associate Professor of Materials Science and Engineering and Wilson Faculty Fellow, presenting “AI-Powered Growth Recipe Optimization”
ABSTRACT: The molecular beam epitaxy systems that Law uses have multiple types of in situ characterization including electron diffraction and spectroscopic ellipsometry. We are interested in developing an AI agent that can view the data in real time and suggest adjustments to growth recipes to optimize film thickness, stoichiometry, and other properties.
Anthony Richardella, assistant research professor, Eberly College of Science, Materials Characterization Lab, presenting “Usage of LLMs for data analysis and discovery of 2DCC materials data”
ABSTRACT: In its 10 years of operation the 2DCC has amassed datasets on approximately 15,000 unique growths of materials, producing over 20,000 samples for users internal and external to Penn State. The LiST database stores the associated metadata, synthesis parameters and characterization data for all these samples that is accessible to users via a website and API. While the data is accessible, and has been used for detailed machine learning studies with Wes Reinhart’s group, the 2DCC is seeking ways utilize AI to give users better summaries of the totality data and deeper insights into how to interpret it, in a user-friendly interface. The capabilities of LLMs to summarize patterns and present them in a comprehensible way is expected to significantly accelerate how research is conducted. Methods to make the data more interpretable by LLMs, optimal data structures, and potential topics of collaboration are all areas we are open to discussions.