Schrödinger Bridges
PulseAugur coverage of Schrödinger Bridges — every cluster mentioning Schrödinger Bridges across labs, papers, and developer communities, ranked by signal.
3 day(s) with sentiment data
-
New research links diffusion model training loss to Fisher geometry
Researchers have analyzed the irreducible excess loss in denoising score matching for diffusion models. They found this excess loss is directly related to the Fisher--Rao metric of the conditional endpoint family, integ…
-
New framework Bridge Graphical Models improves generative AI design
Researchers have introduced Bridge Graphical Models (BGMs) as a new framework to analyze and improve continuous-time generative models. BGMs decouple design choices such as endpoint coupling, bridge law, and Markovian p…
-
New library unifies 12 generative models as mean-field games
Researchers have introduced MFGLab, an open-source PyTorch library that unifies twelve continuous-time generative models under a single variational problem framework. This approach treats models like Continuous Normaliz…
-
New framework enables efficient training of reflected Schrödinger Bridges
Researchers have developed a new framework for training reflected Schrödinger Bridges (SBs) that is more efficient than existing methods. This new approach allows reflected SBs to be trained similarly to flow matching, …
-
Deep Learning Improves Lunar Topography Resolution
Researchers have developed a new deep learning method using diffusion-based Schrödinger Bridges (SBs) to improve the resolution of lunar topography models. This approach connects low-resolution topography data with high…
-
FreeBridge model uses Schrödinger Bridges for cellular transition dynamics
Researchers have developed FreeBridge, a new method for modeling cellular transitions using variational Schrödinger Bridges. This approach infers stochastic processes between control and treated cell populations, even w…
-
New AI method speeds up visual navigation for robots
Researchers have developed Rectified Schrödinger Bridge Matching (RSBM), a new framework designed to improve visual navigation for autonomous agents in Embodied AI. RSBM leverages a shared velocity-field structure betwe…
-
New TR-SBTS Framework Enhances Time Series Generation
Researchers have developed Triangular-Reference Schrödinger Bridges for Time Series (TR-SBTS), an advanced framework for time series generation. This method extends existing Schrödinger Bridges by replacing the standard…
-
New entropy method boosts generative model sample quality
Researchers have developed a new method for optimizing the discretization of generative models, aiming to improve sample quality with limited computational resources. This approach, termed conditional-marginal entropy-r…
-
New QDSB method accelerates generative model training
Researchers have introduced Quantized Diffusion Schrödinger Bridges (QDSB), a novel method for learning generative models from unpaired data. QDSB addresses the computational challenges of traditional Schrödinger bridge…
-
MAPF solved via multi-marginal optimal transport and Schrödinger bridges
Researchers have developed a novel approach to solve multi-agent path finding (MAPF) problems by reformulating them as a specific type of multi-marginal optimal transport (MMOT) problem. This method leverages a Markovia…