We build innovation from scratch: our work starts with top-level research and continues trhough system design and deployment.
We believe that scientific rigor and real-world adaptation must coexist.
CVPR'24
Procedural Learning & Anomaly Detection
Online mistake detection in PRocedural EGOcentric videos.
ICLM'24
Uncertainty Estimation & Domain Adaptation
Hyperbolic Active Learning for Semantic Segmentation under Domain Shift.
ICLM'24
LLMs and Long-Term Dependency recognition
Selective Resampling for Expressive State Space Models.
COLM'25
LLMs and Long-Term Dependency recognition
Evaluating Coding LLMs at 1M Context Windows.
ECCV'24
Multimodal Foundation Models
Hyperbolic Learning with Multimodal Large Language Models.
ICLR'25
Multimodal Foundation Models
Compositional Entailment Learning for Hyperbolic Vision-Language Models.
We apply advanced machine learning methods to complex, real-world data, with a strong focus on multimodal learning and representation.
Our work spans perception, learning under uncertainty, and vision-language models, enabling AI systems to reason across multiple data sources and time scales.
We focus on methods that remain robust under domain shift and imperfect data, conditions that define real operational environments.
We design AI at the system level, not as isolated models.
Our work spans the full stack: from data pipelines and learning architectures to deployment constraints, compute efficiency, and long-term maintainability.
We focus on how components interact in real settings — where perception, decision-making, and learning must operate together under latency, reliability, and safety constraints.
This approach allows us to build AI systems that are not only performant in controlled environments, but stable and scalable in production.
We translate cutting-edge research into deployable, real-world AI systems.
Our team works closely with industry partners to move from prototypes to operational systems, ensuring that research outcomes deliver measurable impact beyond the lab.
The result is AI that is grounded in strong scientific foundations while remaining usable, reliable, and aligned with real business and operational needs.