Register here to attend Kedar Hippalgaonkar 's AC seminar taking place in person at 700 University Avenue in the 10th Floor Seminar Room or online via Zoom.
Symmetry, disorder, generative design and experimental synthesis of ‘novel’ inorganic materials
Property-directed generative design is the holy grail of AI for materials discovery. Most real-world inorganic materials have internal symmetry beyond (P1) lattice translation. Symmetry rules that atoms obey play a fundamental role in determining crystals structure. These symmetries form energy configurations, determine stability, and influence key material structural and functional properties such as electrical and thermal conductivity, optical and polarization behaviour, and mechanical strength. Similarly, compositional disorder is critical, and perhaps, ubiquitous when materials are synthesized. And yet, despite recent advancements, state-of-the-art generative models struggle to generate symmetric crystals and/or deal with disorder, sometimes due to lack of the right data. Wyckoff positions are used as the basis for an elegant, compressed, flexible, and discrete structure representation. To model the distribution, a permutation-invariant autoregressive model based on Transformer is developed, along with a Variational Auto Encoder (VAE) and absence of positional encoding. Experiments demonstrate that Wyckoff-based representation achieves the best performance in generating novel, diverse, stable structures conditioned on the symmetry space group, while also showing competitive metric values compared to models not conditioned on symmetry.
Explore what high-throughput synthesis of inorganic compounds entails, including innovations in hardware that go beyond parallelization. Validation of successful generative design requires careful characterization of structure and composition, with early forays into 'novel' compounds examining not only their synthesis but also their stability against competing phases and their functional properties. Ultimately, achieving true property-directed generative design relies on a closed-loop integration of data-driven computation, modeling, and experiments.
Associate Professor Kedar Hippalgaonkar’s research interests are in AI-driven solid-state materials-by-design. He holds a joint appointment as an associate professor with the Materials Science and Engineering Department at NTU, and as a Principal Scientist at IMRE, A*STAR. He was the Scientific Director of the Multi-PI S$25M Accelerated Materials Development for Manufacturing (AMDM) program from 2018 – 2024, and S$10M Materials Generative Design and Testing Framework program (Mat-GDT) from 2024-2027. Leading a group of >30 members, he has demonstrated clear areas of advancement in the discovery of new functional materials, AI and robotics for accelerated materials discovery, and advancing fundamental knowledge in inequilibrium charge and phonon scattering. His scientific contributions in the materials-by-design space have established a framework for the rapid discovery of materials and new physics, which is now being utilised globally in data-driven research. His commitment to translating scientific research into tangible real-world applications is exemplified by his role as the Co-founder and Senior Scientific Advisor of a startup – Xinterra, Inc. As a contributing member of the Acceleration Consortium at the University of Toronto, Kedar collaborates with an international community of scientists dedicated to the creation of self-driving labs. These platforms are pivotal in unlocking new discoveries in molecules and materials, further expanding the horizon of scientific understanding.
About the AC Seminar Series
The Acceleration Consortium (AC) seminar series explores perspectives on the future of AI for science, presents cutting-edge research findings, enables collaborations, and offers training and upskilling opportunities. Presented both in-person and online, these seminars will host a diverse set of speakers on topics related to accelerated discovery across three tracks:
AC Distinguished Seminars: Leaders in the autonomous discovery community will share their findings and perspectives that are helping to shape future directions and address key challenges. These will be delivered in a hybrid format at the University of Toronto.
AC Early Career Seminars: Early career researchers will present results from their latest publications, taking a technical dive into findings, methods, and tools. These will be delivered in a hybrid format at the University of Toronto.
AC Virtual Training Seminars: Instructors will provide standalone introduction lectures and hands-on tutorials on topics related to self-driving labs with an emphasis on principles, literacy, and skills. These will be delivered virtually.
Do you have a suggestion for a talk? We welcome your ideas for potential speakers from diverse career stages and backgrounds.
