The Timely Approximations Group, led by Kartic Subr, develops algorithms that accelerate the simulation of physical phenomena to enable both design and reactive applications. High-fidelity simulations are often too computationally expensive for interactive use. Our research focuses on creating fast approximations that preserve essential accuracy while drastically reducing computation, making them suitable for real-time constraints and the demands of inverse problem solving. In computer graphics, we design methods to approximate light transport, reflectance, and geometry, enabling interactive rendering, realistic appearance modelling, and neural scene representations that respond instantly to user input. In protein engineering and design, we explore efficient surrogates for molecular interactions, allowing rapid iteration and optimization in large combinatorial design spaces. These approximations enable machine learning–guided workflows that balance speed with predictive power. In robotics, our work focuses on approximating dynamics and control to support agents that must reason and act under tight time budgets in unpredictable environments.

Our agenda unifies these domains under a central principle: timeliness matters. By developing scalable, explainable approximations of physical processes, we aim to unlock new capabilities for systems that must react quickly, adapt continuously, and empower human creativity across science, design, and engineering.

Bio