Searcharxiv⌕ Search

arXiv subjects

Michail Loulakis

Publications and source records attributed to Michail Loulakis.

15 recordsLinked to original sources
arXiv2024

Optimal control, viscosity approximation and Arrhenius Law for the shallow lake problem

We prove existence of optimal control for the deterministic and stochastic shallow lake problem without any restrictions on the parameter space and we establish a generalization of the Arrhenius Law in the case of noise-dependent potentials, which naturally arise in control theory problems. We also prove a result about convergence of the derivatives in the viscosity approximation of the value function and use this result to derive the Arrhenius Law for the shallow lake problem.

math.OC↗
arXiv2023

On the optimally controlled stochastic shallow lake

We consider the stochastic control problem of the shallow lake and continue the work of G. T. Kossioris, Loulakis, and Souganidis (2019) in three directions. First, we generalise the characterisation of the value function as the viscosity solution of a well-posed problem to include more general recycling rates. Then, we prove approximate optimality under bounded controls and we establish quantitative estimates. Finally, we implement a convergent and stable numerical scheme for the computation of the value function to investigate properties of the optimally controlled stochastic shallow lake. This approach permits to derive tail asymptotics for the invariant distribution and to extend results of Grass, Kiseleva, and Wagener (2015) beyond the small noise limit.

math.OC↗
arXiv2023

A new approach to generalisation error of machine learning algorithms: Estimates and convergence

In this work we consider a model problem of deep neural learning, namely the learning of a given function when it is assumed that we have access to its point values on a finite set of points. The deep neural network interpolant is the the resulting approximation of f, which is obtained by a typical machine learning algorithm involving a given DNN architecture and an optimisation step, which is assumed to be solved exactly. These are among the simplest regression algorithms based on neural networks. In this work we introduce a new approach to the estimation of the (generalisation) error and to convergence. Our results include (i) estimates of the error without any structural assumption on the neural networks and under mild regularity assumptions on the learning function f (ii) convergence of the approximations to the target function f by only requiring that the neural network spaces have appropriate approximation capability.

stat.ML↗
arXiv2022

Discrete Gradient Flow Approximations of High Dimensional Evolution Partial Differential Equations via Deep Neural Networks

We consider the approximation of initial/boundary value problems involving, possibly high-dimensional, dissipative evolution partial differential equations (PDEs) using a deep neural network framework. More specifically, we first propose discrete gradient flow approximations based on non-standard Dirichlet energies for problems involving essential boundary conditions posed on bounded spatial domains. The imposition of the boundary conditions is realized weakly via non-standard functionals; the latter classically arise in the construction of Galerkin-type numerical methods and are often referred to as "Nitsche-type" methods. Moreover, inspired by the seminal work of Jordan, Kinderleher, and Otto (JKO) \cite{jko}, we consider the second class of discrete gradient flows for special classes of dissipative evolution PDE problems with non-essential boundary conditions. These JKO-type gradient flows are solved via deep neural network approximations. A key, distinct aspect of the proposed methods is that the discretization is constructed via a sequence of residual-type deep neural networks (DNN) corresponding to implicit time-stepping. As a result, a DNN represents the PDE problem solution at each time node. This approach offers several advantages in the training of each DNN. We present a series of numerical experiments which showcase the good performance of Dirichlet-type energy approximations for lower space dimensions and the excellent performance of the JKO-type energies for higher spatial dimensions.

math.NA↗
arXiv2021

Quantum advantage in biometric authentication with single photons

It was recently proposed to use the human visual system's ability to perform efficient photon counting in order to devise a new biometric methodology. The relevant biometric fingerprint is represented by the optical losses light suffers along several different paths from the cornea to the retina. The fingerprint is accessed by interrogating a subject on perceiving or not weak light flashes, containing few tens of photons, so that the subject's visual system works at the threshold of perception, at which regime optical losses play a significant role. Here we show that if instead of weak laser light pulses we use quantum light sources, in particular single photon sources, we obtain a quantum advantage, which translates into a reduction of the resources required to achieve a desired performance. Besides the particular application on biometrics, our work further demonstrates that quantum light sources can provide deeper insights when studying human vision.

quant-ph↗
arXiv2020

A Simple Network of Nodes Moving on the Circle

Two simple Markov processes are examined, one in discrete and one in continuous time, arising from idealized versions of a transmission protocol for mobile, delay-tolerant networks. We consider two independent walkers moving with constant speed on either the discrete or continuous circle, and changing directions at independent geometric (respectively, exponential) times. One of the walkers carries a message that wishes to travel as far and as fast as possible in the clockwise direction. The message stays with its current carrier unless the two walkers meet, the carrier is moving counter-clockwise, and the other walker is moving clockwise. In that case, the message jumps to the other walker. The long-term average clockwise speed of the message is computed. An explicit expression is derived via the solution of an associated boundary value problem in terms of the generator of the underlying Markov process. The average transmission cost is also similarly computed, measured as the long-term number of jumps the message makes per unit time. The tradeoff between speed and cost is examined, as a function of the underlying problem parameters.

cs.IT↗
arXiv2019

Generalized Young measures and the hydrodynamic limit of condensing zero range processes

Condensing zero range processes (ZRPs) are stochastic interacting particle systems that exhibit phase separation with the emergence of a condensate. Standard approaches for deriving a hydrodynamic limit of the density fail in these models, and an effective macroscopic description has not been rigorously established, yet. In this article we prove that the limiting triple $(π,W,σ)$ of the empirical density, the empirical current, and the empirical jump rate of the ZRP satisfies the continuity equation $\partial_tπ=-{\rm{div}}W$ in the sense of distributions. Here $(π_t)_{t\geq 0}$ is a $w^*$-continuous curve of finite non-negative measures on the torus $\mathbb{T}^d$, $σ_t\in H^1(\mathbb{T}^d)$ and $W_t=-\nablaσ_t$ is a vector-valued measure that is absolutely continuous with respect to the Lebesgue measure, for all almost all $t\geq 0$. In order to obtain a closed equation we propose a generalization of Young measures and we prove that for symmetric ZRPs on the torus, the hydrodynamic limit of the density is a generalized Young-measure-valued weak solution $\boldsymbolπ=(\boldsymbolπ_t)_{t\geq 0}$ to a saturated filtration equation $\partial_t\boldsymbolπ=ΔΦ(\boldsymbolπ)$. Furthermore we prove a one-sided two-blocks estimate and we give an equivalent criterion for its validity. Assuming the validity of the two-blocks estimate one obtains the equation $\partial_tπ=ΔΦ(π^{ac})$ for the empirical density, where $π=π^{ac}+π^\perp$ is the Radon-Nikodym decomposition.

math.PR↗
arXiv2018

Metastable Markov chains: from the convergence of the trace to the convergence of the finite-dimensional distributions

We consider continuous-time Markov chains which display a family of wells at the same depth. We provide sufficient conditions which entail the convergence of the finite-dimensional distributions of the order parameter to the ones of a finite state Markov chain. We also show that the state of the process can be represented as a time-dependent convex combination of metastable states, each of which is supported on one well.

math.PR↗
arXiv2017

The Deterministic and Stochastic Shallow Lake Problem

We study the welfare function of the deterministic and stochastic shallow lake problem. We show that the welfare function is the viscosity solution of the associated Bellman equation, we establish several properties including its asymptotic behaviour at infinity and we present a convergent monotone numerical scheme.

math.PR↗
arXiv2016

Metastability in a condensing zero-range process in the thermodynamic limit

Zero-range processes with decreasing jump rates are known to exhibit condensation, where a finite fraction of all particles concentrates on a single lattice site when the total density exceeds a critical value. We study such a process on a one-dimensional lattice with periodic boundary conditions in the thermodynamic limit with fixed, super-critical particle density. We show that the process exhibits metastability with respect to the condensate location, i.e. the suitably accelerated process of the rescaled location converges to a limiting Markov process on the unit torus. This process has stationary, independent increments and the rates are characterized by the scaling limit of capacities of a single random walker on the lattice. Our result extends previous work for fixed lattices and diverging density in [J. Beltran, C. Landim, Probab. Theory Related Fields, 152(3-4):781-807, 2012], and we follow the martingale approach developed there and in subsequent publications. Besides additional technical difficulties in estimating error bounds for transition rates, the thermodynamic limit requires new estimates for equilibration towards a suitably defined distribution in metastable wells, corresponding to a typical set of configurations with a particular condensate location. The total exit rates from individual wells turn out to diverge in the limit, which requires an intermediate regularization step using the symmetries of the process and the regularity of the limit generator. Another important novel contribution is a coupling construction to provide a uniform bound on the exit rates from metastable wells, which is of a general nature and can be adapted to other models.

math.PR↗
arXiv2015

Cross-Layer Design of Wireless Multihop Networks over Stochastic Channels with Time-Varying Statistics

Network Utility Maximization (NUM) is often applied for the cross-layer design of wireless networks considering known wireless channels. However, realistic wireless channel capacities are stochastic bearing time-varying statistics, necessitating the redesign and solution of NUM problems to capture such effects. Based on NUM theory we develop a framework for scheduling, routing, congestion and power control in wireless multihop networks that considers stochastic Long or Short Term Fading wireless channels. Specifically, the wireless channel is modeled via stochastic differential equations alleviating several assumptions that exist in state-of-the-art channel modeling within the NUM framework such as the finite number of states or the stationarity. Our consideration of wireless channel modeling leads to a NUM problem formulation that accommodates non-convex and time-varying utilities. We consider both cases of non orthogonal and orthogonal access of users to the medium. In the first case, scheduling is performed via power control, while the latter separates scheduling and power control and the role of power control is to further increase users' optimal utility by exploiting random reductions of the stochastic channel power loss while also considering energy efficiency. Finally, numerical results evaluate the performance and operation of the proposed approach and study the impact of several involved parameters on convergence.

eess.SY↗
arXiv2011

Zero Range Condensation at Criticality

Zero-range processes with decreasing jump rates exhibit a condensation transition, where a positive fraction of all particles condenses on a single lattice site when the total density exceeds a critical value. We study the onset of condensation, i.e. the behaviour of the maximum occupation number after adding or subtracting a subextensive excess mass of particles at the critical density. We establish a law of large numbers for the excess mass fraction in the maximum, which turns out to jump from zero to a positive value at a critical scale. Our results also include distributional limits for the fluctuations of the maximum, which change from standard extreme value statistics to Gaussian when the density crosses the critical point. Fluctuations in the bulk are also covered, showing that the mass outside the maximum is distributed homogeneously. In summary, we identify the detailed behaviour at the critical scale including sub-leading terms, which provides a full understanding of the crossover from sub- to supercritical behaviour.

math.PR↗
arXiv2008

Thermodynamic Limit for the Invariant Measures in Supercritical Zero Range Processes

We prove a strong form of the equivalence of ensembles for the invariant measures of zero range processes conditioned to a supercritical density of particles. It is known that in this case there is a single site that accomodates a macroscopically large number of the particles in the system. We show that in the thermodynamic limit the rest of the sites have joint distribution equal to the grand canonical measure at the critical density. This improves the result of Großkinsky, Schütz and Spohn, where convergence is obtained for the finite dimensional marginals. We obtain as corollaries limit theorems for the order statistics of the components and for the fluctuations of the bulk.

math.PR↗