Universal induction relies on some general search procedure that is doomed to be inefficient. One possibility to achieve both generality and efficiency is to specialize this procedure w.r.t. any given narrow task. However, complete specialization that implies direct mapping from the task parameters to solutions (discriminative models) without search is not always possible. In this paper, partial specialization of general search is considered in the form of genetic algorithms (GAs) with a specialized crossover operator. We perform a feasibility study of this idea implementing such an operator in the form of a deep feedforward neural network. GAs with trainable crossover operators are compared with the result of complete specialization, which is also represented as a deep neural network. Experimental results show that specialized GAs can be more efficient than both general GAs and discriminative models.

Alexey Potapov Sergey Rodionov

Person re-identification (Re-ID) is the task of matching humans across cameras with non-overlapping views that has important applications in visual surveillance. Like other computer vision tasks, this task has gained much with the utilization of deep learning methods. However, existing solutions based on deep learning are usually trained and tested on samples taken from same datasets, while in practice one need to deploy Re-ID systems for new sets of cameras for which labeled data is unavailable. Here, we mitigate this problem for one state-of-the-art model, namely, metric embedding trained with the use of the triplet loss function, although our results can be extended to other models. The contribution of our work consists in developing a method of training the model on multiple datasets, and a method for its online practically unsupervised fine-tuning. These methods yield up to 19.1% improvement in Rank-1 score in the cross-dataset evaluation.

Sergey Rodionov Alexey Potapov Hugo Latapie Enzo Fenoglio Maxim Peterson

A new method of regime shift detection in the correlation coefficient is proposed. The method is designed to find multiple change-points with unknown locations in time series. It signals a possible regime shift in real time and allows for its monitoring. The method is tested on randomly generated time series with predefined change-points. It is applied to examine structural changes in the Bering Sea climate. A major shift is found in 1967, which coincides with a transition from a zonal type of atmospheric circulation to a meridional one. The roles of the Siberian and Alaskan centers of action on winter temperatures in the eastern Bering Sea have been investigated.

Sergei Rodionov

Two methods for detecting abrupt shifts in the variance, Integrated Cumulative Sum of Squares (ICSS) and Sequential Regime Shift Detector (SRSD), have been compared on both synthetic and observed time series. In Monte Carlo experiments, SRSD outperformed ICSS in the overwhelming majority of the modelled scenarios with different sequences of variance regimes. The SRSD advantage was particularly apparent in the case of outliers in the series. When tested on climatic time series, in most cases both methods detected the same change points in the longer series (252-787 monthly values). The only exception was the Arctic Ocean SST series, when ICSS found one extra change point that appeared to be spurious. As for the shorter time series (66-136 yearly values), ICSS failed to detect any change points even when the variance doubled or tripled from one regime to another. For these time series, SRSD is recommended. Interestingly, all the climatic time series tested, from the Arctic to the Tropics, had one thing in common: the last shift detected in each of these series was toward a high-variance regime. This is consistent with other findings of increased climate variability in recent decades.

Sergei Rodionov

This paper is a survey on universal algorithms for solving the matrix Bellman equations over semirings and especially tropical and idempotent semirings. However, original algorithms are also presented. Some applications and software implementations are discussed.

Grigory L. Litvinov Anatoly Ya. Rodionov Sergei N. Sergeev Andrei N. Sobolevski

A criterion for the choice of optimal softening length $\epsilon$ and time-step $dt$ for $N$-body simulations of a collisionless stellar system is analyzed. Plummer and Hernquist spheres are used as models to follow how changes in various parameters of an initially equilibrium stable model depend on $\epsilon$ and $dt$. These dependences are used to derive a criterion for choosing $\epsilon$ and $dt$. The resulting criterion is compared to Merritt's criterion for choosing the softening length, which is based on minimizing the mean irregular force acting on a particle with unit mass. Our criterion for choosing $\epsilon$ and $dt$ indicate that $\epsilon$ must be a factor of 1.5-2 smaller than the mean distance between particles in the densest regions to be resolved. The time-step must always be adjusted to the chosen $\epsilon$ (the particle must, on average, travel a distance smaller than $0.5\epsilon$ during one time-step). An algorithm for solving N-body problems with adaptive variations of the softening length is discussed in connection with the task of choosing $\epsilon$, but is found not to be promising.

S. A. Rodionov N. Ya. Sotnikova

We analyze the relationship between the mass of a spherical component and the minimum possible thickness of stable stellar disks. This relationship for real galaxies allows the lower limit on the dark halo mass to be estimated (the thinner the stable stellar disk is, the more massive the dark halo must be). In our analysis, we use both theoretical relations and numerical N-body simulations of the dynamical evolution of thin disks in the presence of spherical components with different density profiles and different masses. We conclude that the theoretical relationship between the thickness of disk galaxies and the mass of their spherical components is a lower envelope for the model data points. We recommend using this theoretical relationship to estimate the lower limit for the dark halo mass in galaxies. The estimate obtained turns out to be weak. Even for the thinnest galaxies, the dark halo mass within four exponential disk scale lengths must be more than one stellar disk mass.

N. Ya. Sotnikova S. A. Rodionov

Aims: We aim to develop an algorithm for constructing equilibrium initial conditions for simulations of disk galaxies with a triaxial halo and/or a gaseous component. This will pave the way for N-body simulations of realistic disk galaxies. Methods: We use the iterative method, which we presented in a previous article. The idea of this method is very simple. It relies on constrained evolution. Results: We develop an algorithm for constructing equilibrium models of disk galaxies including a gaseous disk and a triaxial or axisymmetric halo. We discuss two test models. The first model consists of a spherical halo, a stellar disk, and an isothermal gaseous disk. The second model consists of a triaxial halo, a stellar disk, and a star-forming gaseous disk. We demonstrate that both test models are very close to equilibrium, as we had intended.

S. A. Rodionov E. Athanassoula

We demonstrate that in N-body simulations of isolated disc galaxies there is numerical vertical heating which slowly increases the vertical velocity dispersion and the disc thickness. Even for models with over a million particles in a disc, this heating can be significant. Such an effect is just the same as in numerical experiments by Sellwood (2013). We also show that in a stellar disc, outside a boxy/peanut bulge, if it presents, the saturation level of the bending instability is rather close to the value predicted by the linear theory. We pay attention to the fact that the bending instability develops and decays very fast, so it couldn't play any role in secular vertical heating. However the bending instability defines the minimal value of the ratio between the vertical and radial velocity dispersions $\sigma_z / \sigma_R \approx 0.3$ (so indirectly the minimal thickness) which could have stellar discs in real galaxies. We demonstrate that observations confirm last statement.

S. A. Rodionov N. Ya. Sotnikova

The studies of chemical processes in spatially confined conditions are of interest from the fundamental and industrial points of view. By means of the W band EPR and 1H Mims electron nuclear double resonance (ENDOR) we show that the radiation-induced paramagnetic centers (E') in the synthetic nanoporous silica opals could be used as sensitive probes to investigate the surface modification and, potentially, reactions of polymerization in the confined by opal pores.

Andrei Galukhin Yuri Osin Alexander Rodionov Georgy Mamin Marat Gafurov Sergei Orlinskii