From December 10–16, EDI lead researcher Kārlis Freivalds and researchers Laura Leja and Oskars Teikmanis participated in the Neural Information Processing Systems (NeurIPS 2024) conference in Vancouver, Canada.
The conference offered numerous opportunities for listening to thematic talks, debates, panel discussions, and presentations of submitted papers on a wide range of topics related to machine learning: large language models, object detection, 3D pose estimation, diffusion models, reinforcement learning, and many more. Kārlis, Laura, and Oskars presented two papers:
- “Learning Fluid-Directed Rigid-Body Control” on a device being developed at EDI that could make it possible to move objects using controlled airflows. Such technology could potentially be used for moving fragile objects as well as creating kinetic art.
- “Shaping Flames with Differentiable Physics Simulations” on a simulation-based method for controlled shaping of fire into various forms.
These papers were developed with funding from the VPP MOTE project
