Registered Data

[CT076]

  • Session Time & Room
    • CT076 (1/1) : 3D @E502 [Chair: Evelyn Buckwar]
  • Classification
    • CT076 (1/1) : Stochastic analysis (60H) / Computer science (68-)

[02343] Construction and analysis of splitting methods for Chemical Langevin Equations

  • Session Time & Room : 3D (Aug.23, 15:30-17:10) @E502
  • Type : Contributed Talk
  • Abstract : Consider modeling the stochastic dynamics underlying different chemical systems, which is usually described by the Gillespie Stochastic Simulation Algorithm (SSA), i.e. the Markov process arising from taking into account every single chemical reaction event. While exact and easy to implement, this algorithm is computationally expensive for chemical reactions involving a large number of molecular species. As an approximation, Chemical Langevin Equations (CLEs) can work for large number of species or/and reactions. In this talk, we construct an explicit splitting method applied to the system of CLEs for a simple example of a reversible bimolecular reaction. The drift term of this stochastic differential equation system satisfies a local one-sided Lipschitz condition and the diffusion term involves square root terms. We then present the main ideas of a mean-square convergence proof, as well as numerical illustrations. The results are joint work with Youssra Souli, Johannes Kepler University, Linz.
  • Classification : 60H10, 65C30, 60H35
  • Format : Talk at Waseda University
  • Author(s) :
    • Evelyn Buckwar (Johannes Kepler University)
    • Youssra Souli (Johannes Kepler University)

[02547] Split S-ROCK methods for stiff It\^{o} stochastic differential equations

  • Session Time & Room : 3D (Aug.23, 15:30-17:10) @E502
  • Type : Contributed Talk
  • Abstract : We propose explicit stochastic Runge--Kutta methods for stiff It\^{o} stochastic differential equations. The family of the methods is constructed on the basis of the Runge--Kutta--Chebyshev methods, and we utilize a Strang splitting-type approach. The derived methods achieve weak order $2$, and have high computational accuracy for relatively large time-step size, as well as good stability properties. In numerical experiments, we confirm that our methods are superior to existing methods in computational accuracy.
  • Classification : 60H10, 65L05, 65L06
  • Format : Talk at Waseda University
  • Author(s) :
    • Yoshio Komori (Kyushu Institute of Technology)
    • David Cohen (Chalmers University of Technology)
    • Kevin Burrage (Queensland University of Technology)

[01044] Nonlinear SPDE models of particle systems

  • Session Time & Room : 3D (Aug.23, 15:30-17:10) @E502
  • Type : Contributed Talk
  • Abstract : Interacting particle systems provide flexible and powerful models that are useful in many application areas. However, particle systems with large numbers of particles are very complex. Therefore, a common strategy is to derive effective equations that describe the time evolution of the empirical particle density. Our aim is to consider non-Gaussian models that provide approximation of the Dean-Kawasaki equation. This is the joint work with Kremp and Perkowski.
  • Classification : 60H15, 35Q83, 65M08
  • Format : Talk at Waseda University
  • Author(s) :
    • Ana Djurdjevac (Freie Universität Berlin)

[01806] Well-posedness of a class of SPDE with fully monotone coefficients perturbed by Levy noise

  • Session Time & Room : 3D (Aug.23, 15:30-17:10) @E502
  • Type : Contributed Talk
  • Abstract : In this talk, we consider a class of stochastic partial differential equations with fully locally monotone coefficients in a Gelfand triplet. Under certain generic assumptions of the coefficients, we prove the existence of a probabilistic weak solution as well as the pathwise uniqueness of the solution, which implies the existence of a unique probabilistic strong solution. Finally, we allow both the diffusion and jump noise coefficients to depend on the gradient of the solution.
  • Classification : 60H15, 35R60, 35Q35
  • Format : Talk at Waseda University
  • Author(s) :
    • Ankit Kumar (Indian Institute of Technology, Roorkee, Uttarakhand )
    • Manil T. Mohan (Indian Institute of Technology, Roorkee, Uttarakhand)

[01017] Feature Collisions in Neural Networks: Theory and Practice

  • Session Time & Room : 3D (Aug.23, 15:30-17:10) @E502
  • Type : Contributed Talk
  • Abstract : Deep neural networks are behind many breakthroughs in the last decade, but much of their behavior remains poorly understood. In particular, under some conditions, neural networks can be insensitive to changes of large magnitude, in which case the features are said to collide. We will discuss necessary conditions for such feature collisions to occur, and we will introduce the null-space method, a numerical approach to create data points with colliding features for many vision tasks.
  • Classification : 68-XX, 68Txx, 68T30, 68T07
  • Author(s) :
    • Utku Ozbulak (Ghent University)
    • Joris Vankerschaver (Ghent University)