Omniscale Intelligence Initiative

Mission Statement

At OmniScale AI, our mission is to pioneer the integration of scientific principles from diverse domains through the development of advanced multiscale foundation AI models. Our vision is to transcend traditional boundaries of science, creating a unified framework that bridges scales from the quantum to the cosmic, thereby unlocking unprecedented insights and innovations.

Significance in Material Science:

• Enhanced Predictive Capabilities: We expect multiscale models to enable accurate predictions of material behaviors from the atomic level up to the macroscopic scale, facilitating the design of novel materials with desirable properties.
• Interdisciplinary Insights: By unifying principles from physics, chemistry, and engineering, these models help uncover fundamental mechanisms that govern material performance and reliability.

Impact in Artificial Intelligence:

• Advanced Learning Algorithms: Incorporating multiscale data enhances the robustness and accuracy of AI algorithms, allowing for more nuanced and precise understanding of complex systems.
• Cross-Domain Applicability: These models extend the applicability of AI, enabling it to tackle challenges across diverse scientific and technological fields, from quantum computing to space exploration.

OmniScale AI stands at the frontier of this transformative approach, fostering collaboration and innovation to drive the next generation of intelligent, interconnected scientific research. Join us in our journey to harmonize the scales of knowledge and redefine the future of discovery.

• Konstantin Novoselov
• Andrey Ustyuzhanin
• Kedar Hippalgaonkar
• Lee Wee Sun
• Li Qianxiao
• Nikita Kazeev
• Artem Maevskiy

Join Us

At OmniScale Intelligence, we are dedicated to fostering an inclusive and collaborative environment where visionary leaders and experts from various scientific domains can come together to advance multiscale foundation AI models. If you are passionate about pushing the boundaries of scientific research and innovation, we invite you to join our dynamic think tank.

To express your interest in joining OmniScale AI, please fill out the following form:

• Join OmniScale AI

For more information or any inquiries, please feel free to contact us through the Contact section of our website.

Multiscale Modelling and Machine Learning

We are pleased to invite you to our seminar to foster discussion and collaboration related to Multiscale Modeling. It will be mostly related to materials, life and quantum sciences.

Time: Friday, 2:00 PM – 4:00 PM, bi-weekly

Venue: S17-0406, S17, NUS, Singapore

Upcoming: 2025-02-14

• Speaker: Prof. Duane Loh.

  • Date: Friday, 14 Feb, 2025

  • Time: 3:00 PM – 5:00 PM

  • Venue: S17-04-06

  • Title: Discovering the Secrets of Complexity with AI-companions in Science

  • Abstract: Complex systems are among the most challenging to study due to their intricate interplay of numerous components, encompassing the rich dynamics of many interacting particles, chemical reactions, biological processes, and pharmacological interactions. These systems reflect the complexity of life and the physical world, making their study vital across all branches of science.

    Complex systems are challenging to model because they involve a vast number of interacting degrees of freedom. Capturing and predicting these interactions with high accuracy is daunting due to the sheer scale and complexity of behaviors and phases that can emerge. This diversity makes it difficult to create comprehensive models that can reliably predict system behavior across different scenarios.

    Physicists often approach the challenge of complex systems through a technique known as coarse-graining. This method involves selectively simplifying a system by focusing on key variables while averaging out less critical internal degrees of freedom. By sacrificing some detail, this approach allows for greater predictability and understanding of the system’s behavior at larger length and timescales, as well as at higher levels of complexity.

    Traditionally, coarse-graining has been an art, guided by the intuition and insight of experienced researchers. While this approach can yield powerful results, it often relies on strokes of genius and a fair share of luck. Now, machine learning is democratizing this process, equipping researchers with the tools to identify optimal coarse-graining strategies more efficiently and effectively. This technological advance empowers even the non-expert to uncover insights that were once accessible only to the most seasoned scientists.

    In this talk, I will share the experience of how our students and postdocs, equipped with machine learning tools, have embarked on such journeys of data-driven discovery in complex systems. From uncovering novel spatiotemporal motifs in quantum materials, non-reciprocal interactions between biological cells, deciphering the language of functional disorder in piezoelectric materials, unveiling the structural origins of vibrant colors in butterfly wing scales, to creating novel computational lenses for cryo-electron microscopy. They have collectively pushed the boundaries of how we think about complex systems.

    These experiences, I believe, show us how AI has become indispensable for understanding and discovering the secrets in complex systems.

  • Short biography Duane Loh is an Associate Professor in the Departments of Physics and Biological Sciences at the National University of Singapore (NUS) and a Principal Investigator at the NUS Centre for Bio-imaging Sciences. His research focuses on developing computational lenses, innovative tools that fuse machine learning with scientific and instrument priors to decode complex and chaotic dynamics at the nanometer scale.

    Duane’s group pioneered these computational lenses for single-particle diffractive imaging using X-ray free-electron lasers, where they applied unsupervised learning to discover transient intermediate states and spontaneous order formation in highly dynamic systems. They extended these methods to electron-based imaging, overcoming challenges too complex for traditional hardware-based microscopy alone.

    By combining advanced microscopy with statistical learning, Duane is leading efforts to explore nucleation processes, nanocrystal growth, and self-organization in both physical and biological systems. Building on their expertise in understanding complexity in physical systems,

    Duane’s group is now applying these data-driven approaches to understanding the complex many-body dynamics of biological cells and the spread of vector-borne diseases. This work continues to bridge the gap between massive, complex datasets and foundational scientific understanding, pushing the boundaries of discovery in both physical and biological sciences.

2025-01-31

• Speaker: Prof. Lei Huan,
  • Time: 9:30 AM – 10:30 AM (please note the change of time for this session)
  • Zoom Link: https://nus-sg.zoom.us/j/87589685476?pwd=M99AQERg8SGwJYnzg1gIxVZEmGkBjI.1 (please note that it is an online zoom session)
  • Meeting ID: 875 8968 5476
  • Passcode: 791821
  • Title: An energy-stable machine-learning model of non-Newtonian hydrodynamics with molecular fidelity
  • Abstract: One essential challenge in the computational modeling of multiscale systems is the availability of reliable and interpretable closures that faithfully encode the micro-dynamics. For systems without clear scale separation, there generally exists no such a simple set of macro-scale field variables that allow us to project and predict the dynamics in a self-determined way. We introduce a machine-learning-based approach that enables us to systematically pass the micro-scale physical laws onto the macro-scale. The non-Newtonian hydrodynamics of polymeric fluids is used as an example to illustrate the essential idea. To faithfully retain molecular fidelity, we establish a micro-macro correspondence via a set of encoders for the micro-scale polymer configurations and their macro-scale counterparts, a set of nonlinear conformation tensors. The dynamics of these conformation tensors, with a new form of the objective tensor derivative, can be derived from the micro-scale model and preserves a macro-scale energy variational formulation, and therefore, ensures the energy stability. Unlike our conventional wisdom about ML modeling, the training only uses time-discrete samples. The final model, named the deep non-Newtonian model (DeePN2), retains a multi-scale nature with clear physical interpretation and strictly preserves the frame-indifference constraints. We show that DeePN2 can faithfully capture the broadly overlooked viscoelastic differences arising from the specific molecular structural mechanics without human intervention.

2025-01-17

• Zhu Aiqing, review some insights and results from Weinan E’s book, Principles of Multiscale Modeling
• Chen Mengyi, review of Wang J. et al. Machine learning of coarse-grained molecular dynamics force fields – 2019