Source author record

Vladimir V. Palyulin

Vladimir V. Palyulin appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

cond-mat.stat-mech Biological Physics cond-mat.dis-nn physics.comp-ph cond-mat.soft Quantitative Methods

Catalog footprint

What is connected

5works

6topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2022arXiv

Gradient dynamics in reinforcement learning

Despite the success achieved by the analysis of supervised learning algorithms in the framework of statistical mechanics, reinforcement learning has remained largely untouched. Here we move towards closing the gap by analyzing the dynamics of the policy gradient algorithm. For a convex problem, we show that it obeys a drift-diffusion motion with coeffcients tuned by learning rate. Furthermore, we propose a mapping between a non-convex reinforcement learning problem and a disordered system. This mapping enables us to show how the learning rate acts as an effective temperature and thus is capable of smoothing rough landscapes, corroborating what is displayed by the drift-diffusive description and paving the way for physics-inspired algorithmic optimization based on annealing procedures in disordered systems.

preprint2020arXiv

Scaling up the lattice dynamics of amorphous materials by orders of magnitude

We generalise the non-affine theory of viscoelasticity for use with large, well-sampled systems of arbitrary chemical complexity. Having in mind predictions of mechanical and vibrational properties of amorphous systems with atomistic resolution, we propose an extension of the Kernel Polynomial Method (KPM) for the computation of the vibrational density of states (VDOS) and the eigenmodes, including the $Γ$-correlator of the affine force-field, which is a key ingredient of lattice-dynamic calculations of viscoelasticity. We show that the results converge well to the solution obtained by direct diagonalization (DD) of the Hessian (dynamical) matrix. As is well known, the DD approach has prohibitively high computational requirements for systems with $N=10^4$ atoms or larger. Instead, the KPM approach developed here allows one to scale up lattice dynamic calculations of real materials up to $10^6$ atoms, with a hugely more favorable (linear) scaling of computation time and memory consumption with $N$.

preprint2014arXiv

Optimization of random search processes in the presence of an external bias

We study the efficiency of random search processes based on L{é}vy flights with power-law distributed jump lengths in the presence of an external drift, for instance, an underwater current, an airflow, or simply the bias of the searcher based on prior experience. While Lévy flights turn out to be efficient search processes when relative to the starting point the target is upstream, in the downstream scenario regular Brownian motion turns out to be advantageous. This is caused by the occurrence of leapovers of L{é}vy flights, due to which L{é}vy flights typically overshoot a point or small interval. Extending our recent work on biased LF search [V. V. Palyulin, A. V. Chechkin, and R. Metzler, Proc. Natl. Acad. Sci. USA, DOI:10.1073/pnas.1320424111] we establish criteria when the combination of the external stream and the initial distance between the starting point and the target favors L{é}vy flights over regular Brownian search. Contrary to the common belief that L{é}vy flights with a L{é}vy index $α=1$ (i.e., Cauchy flights) are optimal for sparse targets, we find that the optimal value for $α$ may range in the entire interval $(1,2)$ and include Brownian motion as the overall most efficient search strategy.

preprint2013arXiv

Searching with and against the stream: Levy or Brown?

We study the efficiency of search processes based on Levy flights (LFs) with power-law distributed jump lengths in the presence of an external drift. While LFs turn out to be efficient search processes when relative to the starting point the target is upstream, in the downstream scenario regular Brownian motion turns out to be advantageous. This is caused by the occurrence of leapovers of LFs, due to which LFs typically overshoot a point in space. We establish criteria when the combination of the external stream and the initial distance between the starting point and the target favors LFs over regular Brownian search. Contrary to the common belief that LFs with a stable index alpha=1 are optimal, we find that the optimal alpha may range in the entire interval (1,2) and even include Brownian motion as the overall most efficient search strategy.

preprint2012arXiv

How a finite potential barrier decreases the mean first passage time

We consider the mean first passage time of a random walker moving in a potential landscape on a finite interval, starting and end points being at different potentials. From analytical calculations and Monte Carlo simulations we demonstrate that the mean first passage time for a piecewise linear curve between these two points is minimised by introduction of a potential barrier. Due to thermal fluctuations this barrier may be crossed. It turns out that the corresponding expense for this activation is less severe than the gain from an increased slope towards the end point. In particular, the resulting mean first passage time is shorter than for a linear potential drop between the two points.