A unified diffusion-based particle framework for stochastic optimisation with intractable gradients, with exponential contraction and finite-particle error bounds, instantiated as new momentum and higher-order Langevin algorithms.
The gradient is given for allx∈R d by ∇xEλ(θ,x) =∇ xg1(x) + 1 λ(x−prox λ g2(θ,x)), 27 where we treatθ as fixed in the minimisation overRdx
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Particle-based Generalised Stochastic Optimisation
A unified diffusion-based particle framework for stochastic optimisation with intractable gradients, with exponential contraction and finite-particle error bounds, instantiated as new momentum and higher-order Langevin algorithms.