Disordered Systems and Neural Networks (cond-mat.dis-nn)

  • PDF
    Data is scaling exponentially in fields ranging from genomics to neuroscience to economics. A central question is whether modern machine learning methods can be applied to construct predictive models based on large data sets drawn from complex, natural systems like cells and brains. In machine learning, the predictive power or generalizability of a model is determined by the statistics of training data. In this paper, we ask how predictive inference is impacted when training data is generated by the statistical behavior of a physical system. We develop an information-theoretic analysis of a canonical problem, spin network inference. Our analysis reveals the essential role that thermal fluctuations play in determining the efficiency of predictive inference. Thermal noise drives a system to explore a range of configurations providing `raw' information for a learning algorithm to construct a predictive model. Conversely, thermal energy degrades information by blurring energetic differences between network states. In general, spin networks have an intrinsic optimal temperature at which inference becomes maximally efficient. Simple active learning protocols allow optimization of network temperature, without prior knowledge, to dramatically increase the efficiency of inference. Our results reveal a fundamental link between physics and information and show how the physical environment can be tuned to optimize the efficiency of machine learning.
  • PDF
    We argue that the elastic backbone (EB) (union of shortest paths) on a cylindrical system, recently studied by Sampaio Filho et al. [Phys. Rev. Lett. 120, 175701 (2018)], is in fact the backbone of two-dimensional directed percolation (DP). We simulate the EB on the same system as considered by these authors, and also study the DP backbone directly using an algorithm that allows backbones to be generated in a completely periodic manner. We find that both the EB in the bulk and the DP backbone have a fractal dimension of $d_{b} = d_{B,\rm DP} = 1.681\,02(15)$ at the identical critical point $p_{c,\rm{DP}} \approx 0.705\,485\,22$. We also measure the fractal dimension at the edge of the EB system and for the full DP clusters, and find $d_e = d_{\rm DP} = 1.840 \, 54 (4)$. We argue that those two fractal dimensions follow from the DP exponents as $d_{B,\rm DP} = 2-2\beta/\nu_\parallel = 1.681 \, 07 2 (12) $ and $d_{\rm DP} = 2-\beta/\nu_\parallel = 1.840\, 536 (6)$. Our fractal dimensions differ from the value 1.750(3) found by Sampaio Filho et al., whose value may represent a crossover effect.
  • PDF
    Granular crystallisation is an important phenomenon whereby ordered packing structures form in granular matter under vibration. However, compared with the well-developed principles of crystallisation at the atomic scale, crystallisation in granular matter remains relatively poorly understood. To investigate this behaviour further and bridge the fields of granular matter and materials science, we simulated mono-disperse spheres confined in cylindrical containers to study their structural dynamics during vibration. By applying adequate vibration, disorder-to-order transitions were induced. Such transitions were characterised at the particle scale through bond orientation order parameters. As a result, emergent crystallisation was indicated by the enhancement of the local order of individual particles and the number of ordered particles. The observed heterogeneous crystallisation was characterised by the evolution of the spatial distributions via coarse-graining the order index. Crystalline regimes epitaxially grew from templates formed near the container walls during vibration, here termed the wall effect. By varying the geometrical dimensions of cylindrical containers, the obtained crystallised structures were found to differ at the cylindrical wall zone and the planar bottom wall zone. The formed packing structures were quantitatively compared to X-ray tomography results using again these order parameters. The findings here provide a microscopic perspective for developing laws governing structural dynamics in granular matter.
  • PDF
    Power-law-shaped avalanche-size distributions are widely used to probe for critical behavior in many different systems, particularly in neural networks. The definition of avalanche is ambiguous. Usually, theoretical avalanches are defined as the activity between a stimulus and the relaxation to an inactive absorbing state. On the other hand, experimental neuronal avalanches are defined by the activity between consecutive silent states. We claim that the latter definition may be extended to some theoretical models to characterize their power-law avalanches and critical behavior. We study a system in which the separation of driving and relaxation time scales emerges from its structure. We apply both definitions of avalanche to our model. Both yield power-law-distributed avalanches that scale with system size in the critical point as expected. Nevertheless, we find restricted power-law-distributed avalanches outside of the critical region within the experimental procedure, which is not expected by the standard theoretical definition. We remark that these results are dependent on the model details.
  • PDF
    The superfluid to insulator quantum phase transition of a three-dimensional particle-hole symmetric system of disordered bosons is studied. To this end, a site-diluted quantum rotor Hamiltonian is mapped onto a classical (3+1)-dimensional XY model with columnar disorder and analyzed by means of large-scale Monte Carlo simulations. The superfluid-Mott insulator transition of the clean, undiluted system is in the 4D XY universality class and shows mean-field critical behavior with logarithmic corrections. The clean correlation length exponent $\nu = 1/2$ violates the Harris criterion, indicating that disorder must be a relevant perturbation. For nonzero dilutions below the lattice percolation threshold of $p_c = 0.688392$, our simulations yield conventional power-law critical behavior with dilution-independent critical exponents $z=1.67(6)$, $\nu = 0.90(5)$, $\beta/\nu = 1.09(3)$, and $\gamma/\nu = 2.50(3)$. The critical behavior of the transition across the lattice percolation threshold is controlled by the classical percolation exponents. Our results are discussed in the context of a classification of disordered quantum phase transitions, as well as experiments in superfluids, superconductors and magnetic systems.
  • PDF
    The impact of geometry on many body localization is studied on simple, exemplary systems amenable to exact diagonalization treatment. The crossover between ergodic and MBL phase for uniform as well as quasi-random disorder is analyzed using statistics of energy levels. It is observed that the transition to many-body localized phase is correlated with the number of nearest coupled neighbors. The crossover from extended to localized systems is approximately described by the so called plasma model.

Recent comments

Johnnie Gray Feb 01 2018 12:59 UTC

Thought I'd just comment here that we've rather significantly updated this paper.

Travis Scholten Oct 02 2015 03:25 UTC

Apologies for the delayed reply.

No worries with regards to the code - when it does get released, would you mind pinging me? You can find me on [GitHub](https://github.com/Travis-S).

Nicola Pancotti Sep 23 2015 07:58 UTC

Hi Travis

Yes, that code is related to the work we did and that is my repo. However it is quite outdated. I used that repo for sharing the code with my collaborators. Now we are working for providing a human friendly version, commented and possibly optimized. If you would like to have a working

Travis Scholten Sep 21 2015 17:08 UTC

Has anyone found some source code for the SGD referenced in this paper? I came across a [GitHub repository](https://github.com/nicaiola/thesisproject) from Nicola Pancotti (at least, I think that is his username, and the code seems to fit with the kind of work described in the paper!). I am not sure

Salvatore Mandrà Nov 21 2014 13:50 UTC

The manuscript has been widely revised to focus the reader's attention on the proposed method and its application in presence of local disorder.


Salvatore, Gian Giacomo and Alán

Salvatore Mandrà Aug 01 2014 19:11 UTC

Thanks Dr. Hastings for your comment. It is true that the transverse field Ising model does not satisfy the requirements to apply our method with an exponential reduction. Indeed, the opposite would be quite impressive since the random Ising model is a NP-Hard problem and we ourselves would be prett

Matt Hastings Aug 01 2014 16:43 UTC

The "quite general" conditions seem not to include the transverse field Ising model, the subject of most of the intensive numerical work previously. Incidentally, the terms "Lanczos" and "Krylov subspace" might be helpful.