2016 Workshop on
Statistical Physics of Disordered Systems
and Its Applications (SPDSA2016)
--- Statistical-Mechanical Informatics and Statistical Machine Learning Theory in Big Data Sciences ---

Schedule: 27-28 January, 2016

Venue: Meeting Room Horai in Akiu Resort Hotel Sakan (Access)
    Yumoto Akiu, Taihaku-Ku, Sendai 982-0241, Japan

Scientific Program:

      Organized session only: We do not accept submissions from non-invited speakers.

    27 January, 2016

      13:20-13:30 Opening
        Kazuyuki Tanaka (Graduate School of Information Sciences, Tohoku University, Japan)

      13:30-15:00 Session 1
        Hidetoshi Nishimori (Department of Physics, Tokyo Institute of Technology, Japan)
          Title: Effects of thermal noise on the performance of quantum annealing
            Thermal noise is considered to play detrimental roles in quantum annealing. We have analyzed a spin system coupled with a bosonic bath to analytically evaluate the effects of thermal noise, and have found that the noise simply shifts the effective dimension of the system.
        Tommaso Rizzo (Dipartmento di Fisica, Universitá di Roma, "La Sapienza", Italy)
          Title: Loop correntions to the Bethe approximation: theory and applications
            The Bethe approximation and its algorithmic counterpart, the Belief Propagation algorithm, are at the crossroad between different disciplines, ranging from physics, to optimisation, artificial intelligence and computer science. The Kikuchi approximation is a well-known improvement of the Bethe approximation for system of low dimensionality characterised by a huge number of small loops in the factor graph. I will discuss a class of algorithms designed to improve systematically the predictions of the Bethe approximation which is especially suited to treat systems that have locally a tree-like structure.

      15:00-15:15 Break

      15:15-17:30 Session 2

        Koji Hukushima (Graduate School of Arts and Sciences, The University of Tokyo, Japan)
          Title: Irreversible Monte Carlo approach to continuous spin systems
            The event-chain Monte Carlo (ECMC) algorithm is rejection-free and realizes an irreversible Markov chain that satisfies global balance. The algorithm was originally developed for a hard-sphere system and recently generalized to finite-temperature systems. Here we apply the ECMC algorithm to the three-dimensional ferromagnetic Heisenberg spin model. The autocorrelation functions of some physical quantities indicate a dynamical critical exponent z ≈ 1 at the critical temperature, which is substantially reduced from the conventional value of z ≈ 2 expected from diffusive dynamics.
        Nial Friel (School of Mathematical Sciences, University College Dublin, Ireland)
          Title: Exploiting multi-core architectures for reduced-variance estimation with intractable likelihoods
            Abstract: Many popular statistical models for complex phenomena are intractable, in the sense that the likelihood function cannot easily be evaluated. Bayesian estimation in this setting remains challenging, with a lack of computational methodology to fully exploit modern processing capabilities. In this paper we introduce novel control variates for intractable likelihoods that can dramatically reduce the Monte Carlo variance of Bayesian estimators. We prove that our control variates are well-defined and provide a positive variance reduction. Furthermore we show how to optimise these control variates for variance reduction. The methodology is highly parallel and offers a route to exploit multi-core processing architectures that complements recent research in this direction. Indeed, our work shows that it may not be necessary to parallelise the sampling process itself in order to harness the potential of massively multi-core architectures. Simulation results presented on the Ising model, exponential random graph models and non-linear stochastic differential equation models support our theoretical findings.
            Reference: Bayesian Analysys (2015), DOI: 10.12.14/I5-BA948
        Masayuki Ohzeki (Graduate School of Informatics, Kyoto University, Japan)
          Title: Stochastic gradient method with accelerated stochastic dynamics
            We propose a novel technique to implement stochastic gradient methods, which are beneficial for learning from large datasets, through accelerated stochastic dynamics. A stochastic gradient method is based on mini-batch learning for reducing the computational cost when the amount of data is large. The stochasticity of the gradient can be mitigated by the injection of Gaussian noise, which yields the stochastic Langevin gradient method; this method can be used for Bayesian posterior sampling. However, the performance of the stochastic Langevin gradient method depends on the mixing rate of the stochastic dynamics. In this study, we propose violating the detailed balance condition to enhance the mixing rate. Recent studies have revealed that violating the detailed balance condition accelerates the convergence to a stationary state and reduces the correlation time between the samplings. We implement this violation of the detailed balance condition in the stochastic gradient Langevin method and test our method for a simple model to demonstrate its performance.
            Reference: arxiv.org/abs/1511.06036

    28 January, 2016
      10:15-12:30 Session 3
        Yingying Xu (Interdisciplinary Graduate School of Science and Engineering, Tokyo Institute of Technology, Japan)
          Title: Statistical mechanics analysis of 1-bit compressed sensing
            The one-bit compressed sensing framework aims to reconstruct a sparse signal by only using the sign information of its linear measurements. To compensate for the loss of scale information, past studies in the area have proposed recovering the signal by imposing an additional constraint on the L2-norm of the signal. Here, we examine an alternative strategy that captures scale information by introducing a threshold parameter to the quantization process. We analyze the typical performance using statistical mechanics methods. We also develop a heuristic that adaptively tunes the threshold parameter based on measurement results. Numerical experiments show that the heuristic exhibits satisfactory performance while incurring low computational cost.
        Ayaka Sakata (Department of Statistical Modeling, The Institute of Statistical Mathematics, Tokyo, Japan)
          Title: Estimation of generalized degrees of freedom for sparse regularization
            For the sparse estimation problems performed as penalized maximum likelihood with sparse regularization, the selection of the regularization parameter corresponds to the model selection. AIC for sparse estimation is proposed using the equivalence between Mallows' Cp and AIC for Gaussian likelihood. In the estimation of the model selection criterion, we need to quantify the generalized degrees of freedom (GDF). In case of Lasso, it is mathematically proven that the number of the non-zero components corresponds to the unbiased estimator of GDF. However, there is no general analytical method to estimate GDF that is applicable to the other sparse regularization. We apply a statistical mechanical method to the estimation of GDF and derive its analytical form.
        Chihiro Nakajima (WPI Advanced Institute for Materials Research, Tohoku University, Japan)
          Title: Sparse reconstruction of 3D atomic arrangement : Toward atomic-scale electron CT
            In present, technical basis for an atomic-scale computer tomography (CT) is established by an advanced use of transmission electron microscopy (TEM). However, it is still difficult to obtain many projected images for single object from various directions since desorption of atoms is caused by irradiation of high-energy electrons. We propose the framework for the 3D-image reconstruction of nano-scale object from a few (three) projected images by a sparse regularization algorithm. In the presentation the reconstruction from experimental data with a gold nano-porous cluster is discussed.
      12:30-12:40 Closing

Organized by

    Masayuki Ohzeki (Graduate School of Informatics, Kyoto University, Japan)
    Kazuyuki Tanaka (Graduate School of Information Sciences, Tohoku University, Japan)

Supported by

    JST-CREST``Foundations of Innovative Algorithms for Big Data''

Co-Supported by

    Graduate School of Information Sciences (GSIS), Tohoku University, Japan

Related Webpages

    SPDSA2013 (20-21 March, 2013, Sendai, Japan)
    SPDSA2014 (10 March, 2014, Kyoto, Japan)
    SPDSA2015 (19 February, 2015, Kyoto, Japan)

Contact to SPDSA2016 office