arXiv Science⌕ Search

arXiv subjects

Ewsey Obzherin

Publications and source records attributed to Ewsey Obzherin.

5 recordsLinked to original sources

Finite-Step Geometry of Marginal-Preserving Velocity Fields in Generative Flows

Generative flows can admit different velocity fields that induce the same continuous marginal evolution but behave differently after numerical discretization. We study this finite-step non-equivalence while keeping the marginal path and numerical method fixed. Projected modified equations characterize the component of local integration error that changes the marginal law and give sufficient conditions for higher marginal convergence order. For affine fixed-shape distributions, we characterize exact finite-step reachability and derive spectral obstructions and sharp nearest-reachable Wasserstein errors. For diffusion probability-flow ODEs, we construct skew-score perturbations that preserve the exact marginal path under exact scores and quantify their leakage under score error. Controlled Gaussian and mixture experiments recover the predicted order, reachability, and obstruction phenomena.

cs.LG↗

Four-Dimensional Semi-Homogeneous Convex Cones: A Complete Classification

We classify all four-dimensional semi-homogeneous convex cones, i.e., closed, pointed, full-dimensional cones whose connected automorphism group has principal orbits of codimension at most one. On such cones, the Monge-Ampère equation defining the canonical barrier reduces to an ordinary differential equation. This makes this barrier computationally accessible. The semi-homogeneous cones can then be used in modeling by conic programs. An obvious subset of such cones are the homogeneous cones. The key point allowing the classification is that for all other cones, the automorphism group has a three-dimensional subgroup acting locally faithfully on the principal orbits. This subgroup must have a Lie algebra which is either abelian or a direct product of a 1-dimensional algebra and a two-dimensional algebra generated by $A,B$ satisfying $[A,B]=B$. In both cases we classify all 4-dimensional representations up to isomorphism, and for each representation all invariant convex cones. For every cone an orbit invariant is computed.

math.DG↗

Density-Level Backward Error for Low-NFE Generative Flows

Low-NFE generative ODE sampling is constrained not only by solver order, but also by how a fixed budget of network evaluations is distributed along the flow. We develop a density-level backward-error framework that identifies the component of a solver's modified-equation defect that changes the evolving marginal, while excluding solenoidal trajectory distortion that is invisible in Wasserstein tangent geometry. The resulting residual induces a solver-order-aware compute-allocation rule and, through gauge freedom, an objective for selecting numerically favorable velocity representatives of the same density path. This separates our approach from schedule search driven by trajectory or endpoint surrogates and from higher-order solver design. Exact Gaussian and mixture models establish the mechanism, including cases where curvature-based allocation fails and continuum-optimal transport is not discretization-optimal. On a pretrained CIFAR-10 EDM probability-flow ODE, density-aware allocation consistently improves fixed-update Euler sampling in the low-NFE regime. Matched-seed convergence to a dense Euler reference confirms that the method accelerates approximation of the same high-NFE flow rather than changing the asymptotic sampler.

cs.LG↗

One-Dimensional Symmetry Reductions of the Monge-Ampère Equation on Some Semi-Homogeneous Convex Cones in $\mathbb{R}^4$

Motivated by non-symmetric conic optimization, we construct six explicit families of semi-homogeneous proper convex cones $K$ in $\mathbb R^4$. For each family, a specified three-dimensional subgroup of linear automorphisms of $K$ has generic orbits of codimension one. This yields an orbit invariant $t$ and reduces the Monge-Ampère equation $\det F''=e^{2F}$, defining the canonical barrier on $K$, to an ordinary differential equation (ODE) with independent variable $t$. The families include asymmetric power, rotational-power, exponential, and nonabelian-type cones. For the power and exponential cases we derive first integrals and parametric solutions of the ODE. The reductions open the way for computing the canonical barrier on these cones numerically.

math.OC↗

Decentralized Optimization with Cross-Coupled Mixed Affine Constraints

We study decentralized optimization with cross-coupled mixed affine constraints, where local and shared variables interact through two affine channels. We show that the intrinsic difficulty of combining separately well-conditioned channels is governed by their Friedrichs angle. This geometry induces a cross-coupling factor that cannot be removed by channelwise preconditioning and governs the additional affine-oracle and communication complexity. We develop an accelerated decentralized method with matching minimax guarantees in the smooth strongly convex regime and extend the framework to smooth and nonsmooth convex objectives. Experiments confirm the predicted dependence on cross-channel geometry and network conditioning.

math.OC↗