Our research combines deep learning with physical constraints, domain expertise, and practical constraints to build AI that actually works in the real world—not just in theory.
127
Publications
4.2K+
Citations
48
Researchers
23
Active Projects
Our research spans multiple domains, combining theoretical advances with practical applications that solve real-world problems.
Embedding physical laws and constraints into neural networks to create models that respect fundamental principles of physics, chemistry, and engineering.
Combining vision, language, audio, and sensor data to build AI systems that understand the world like humans do—through multiple simultaneous channels.
Transferring knowledge across domains, languages, and modalities to solve problems where labeled data is scarce or non-existent.
Moving beyond correlation to understand cause-and-effect relationships, enabling AI systems that can reason, intervene, and predict outcomes.
Developing smaller, faster, more energy-efficient models that can run on edge devices without sacrificing accuracy.
Building AI systems that augment human capabilities, explain their reasoning, and work seamlessly alongside domain experts.
Our researchers publish at top venues including NeurIPS, ICML, ICLR, Nature, and Science. Explore our latest findings.
Chen, L., Rodriguez, M., & Kim, S. et al.
We present a novel approach to turbulence modeling that combines physics-informed neural networks with real-time CFD simulation, achieving 100x speedup over traditional methods while maintaining accuracy within 2% of high-fidelity simulations. This work has direct applications in aircraft design and optimization.
Patel, R. & Anderson, T.
Novel methods for extracting causal relationships from complex manufacturing process data with applications to quality control.
Zhang, Y., Liu, H. & Schmidt, M.
A hierarchical attention network that captures binding dynamics across multiple spatial and temporal scales for drug discovery.
Williams, J. & Okonkwo, A.
Self-training methods that adapt across geographies and crop types with minimal labeled data requirements.
Garcia, E. & Thompson, S.
Layer-wise relevance propagation combined with counterfactual explanations for transparent credit scoring.
Mueller, K. & Singh, P.
Acoustic and vibration-based anomaly detection using contrastive learning for predictive maintenance.
Nakamura, T. & Lee, C.
Efficient transformer architecture for real-time damage detection in bridges, pipelines, and industrial facilities.
We don't just publish papers—we build systems that work. Our research methodology bridges the gap between academic innovation and real-world deployment.
We start by understanding the physical constraints, domain expertise, and practical requirements of each problem.
We develop novel algorithms grounded in rigorous mathematical and physical principles.
We test extensively on benchmark datasets and real-world scenarios from our industry partners.
We optimize models for deployment and work with our experts to implement solutions in the real world.
Basic Research
NeurIPS, ICML, ICLR
Applied Research
Domain Adaptation
Engineering
MLOps, Optimization
Production
Real-World Impact
87%
Research to Production
Our team includes PhDs from top institutions, former researchers from leading tech companies, and domain experts with deep industry experience.
Director of Research
Former Google Brain. PhD MIT. Physics-Informed Neural Networks.
48
Researchers
12
PhD Institutions
8
Industry Former
15
Domain Experts
We're always looking for exceptional researchers and engineers. If you're passionate about pushing the boundaries of AI, we'd love to hear from you.
View Open PositionsWe partner with leading academic institutions, research labs, and industry leaders to advance the state of AI and bring innovations to the real world.
MIT, Stanford, Carnegie Mellon, Berkeley, ETH Zurich, Oxford
Boeing, Pfizer, Goldman Sachs, John Deere, Siemens
OpenAI, DeepMind, Meta AI, Google Research, Microsoft Research
Developing privacy-preserving ML methods that enable cross-institution collaboration without sharing raw patient data.
Express Interest →Creating physics-informed digital twins to optimize carbon capture processes in industrial settings.
Express Interest →Building foundation models for robotics that can generalize across tasks and environments with minimal fine-tuning.
Express Interest →Whether you're interested in research collaboration, academic partnerships, or applying our findings to your industry—we'd love to connect.
Location
Minneapolis, Minnesota
Get exclusive access to our latest publications, research insights, and early preprints before they're published.
Join 5,000+ researchers and practitioners. Unsubscribe anytime.