Optimal Transport
A mathematical theory concerned with moving one probability distribution to another at minimum cost. Originating from Monge's 1781 earth-moving problem and reformulated as a linear program by Kantorovich, its central result is Brenier's theorem, which establishes conditions under which the optimal transport map is the gradient of a convex function.
Timeline (last 90 days)
On a single-step dependency graph with a small number of nodes, using approximation algorithms such as Sinkhorn iteration can reduce the time to solve the Optimal Transport to the millisecond level.
Optimal Transport is a mathematical framework for measuring the difference between two probability distributions, originally proposed by Monge in the 18th century, also called the Earth Mover's Distance.
OnTrack uses Optimal Transport to measure the structural difference between the current trajectory and the reference trajectory, rather than a simple string or embedding similarity comparison.
Wasserstein distance bypasses the label permutation problem by representing clustering partitions as probability measures and solving an optimal transport problem, measuring the minimum total cost required to transport one partition into another.
DT-DRO converts diagnostic information into outcome-level transport, simultaneously adjusting both the center of the ambiguity set and the underlying cost of transport.
The Wasserstein distance measures the minimum cost required to transport one distribution into another, where the magnitude of the cost is determined by the underlying cost function.
At the local level, IAU-FOA clusters patch tokens into local patterns and performs fine-grained feature matching via optimal transport (OT).
Optimal transport (OT) originates from the earth-moving problem proposed by Monge in the 19th century, with the modern formulation given by Kantorovich, and is used to measure the Wasserstein distance between feature distributions.
The key challenge of particle morphing lies in particle pairing, and industrial-grade implementations often rely on optimal transport or the Hungarian algorithm to minimize particle movement distance.
The presenter arranged various generative algorithms along a spectrum by degree of endpoint dependence: at the far left are flow matching and stochastic interpolants with independent endpoints (simple to train but requiring multi-step integration at inference), and at the far right is full optimal transport (trivially straight paths but difficult to train).
6 more timeline events
All Facts (16)
On a single-step dependency graph with a small number of nodes, using approximation algorithms such as Sinkhorn iteration can reduce the time to solve the Optimal Transport to the millisecond level.
50%UnverifiedOptimal Transport is a mathematical framework for measuring the difference between two probability distributions, originally proposed by Monge in the 18th century, also called the Earth Mover's Distance.
50%UnverifiedOnTrack uses Optimal Transport to measure the structural difference between the current trajectory and the reference trajectory, rather than a simple string or embedding similarity comparison.
50%UnverifiedWasserstein distance bypasses the label permutation problem by representing clustering partitions as probability measures and solving an optimal transport problem, measuring the minimum total cost required to transport one partition into another.
50%UnverifiedDT-DRO converts diagnostic information into outcome-level transport, simultaneously adjusting both the center of the ambiguity set and the underlying cost of transport.
50%UnverifiedThe Wasserstein distance measures the minimum cost required to transport one distribution into another, where the magnitude of the cost is determined by the underlying cost function.
50%UnverifiedAt the local level, IAU-FOA clusters patch tokens into local patterns and performs fine-grained feature matching via optimal transport (OT).
50%UnverifiedOptimal transport (OT) originates from the earth-moving problem proposed by Monge in the 19th century, with the modern formulation given by Kantorovich, and is used to measure the Wasserstein distance between feature distributions.
50%UnverifiedThe key challenge of particle morphing lies in particle pairing, and industrial-grade implementations often rely on optimal transport or the Hungarian algorithm to minimize particle movement distance.
50%UnverifiedThe presenter arranged various generative algorithms along a spectrum by degree of endpoint dependence: at the far left are flow matching and stochastic interpolants with independent endpoints (simple to train but requiring multi-step integration at inference), and at the far right is full optimal transport (trivially straight paths but difficult to train).
50%UnverifiedBrenier's theorem (1991) states that when the source distribution is absolutely continuous, the unique optimal map transporting the source distribution to the target distribution under squared Euclidean cost must be the gradient of some convex function (the Brenier potential).
50%UnverifiedIn the one-dimensional case, the optimal transport map has a closed-form solution that can be expressed using the cumulative distribution function of the source distribution and the quantile function of the target distribution.
50%UnverifiedFrench mathematician Cédric Villani won the 2010 Fields Medal for his research in the field of optimal transport.
50%UnverifiedSoviet mathematician Leonid Kantorovich reformulated and generalized optimal transport in the 1940s, work that later earned him the Nobel Prize in Economics.
50%UnverifiedStippling generation is essentially an application scenario of Optimal Transport theory.
50%UnverifiedA developer shared on Reddit an experiment in stippling generation based on the idea of optimal transport, transforming ordinary images into works of art using 20,000 dots.
50%