Federated learning is a distributed machine learning paradigm that trains a global model for prediction based on a number of local models at clients while local data privacy is preserved. Class imbalance is believed to be one of the factors that degr
ades the global model performance. However, there has been very little research on if and how class imbalance can affect the global performance. class imbalance in federated learning is much more complex than that in traditional non-distributed machine learning, due to different class imbalance situations at local clients. Class imbalance needs to be re-defined in distributed learning environments. In this paper, first, we propose two new metrics to define class imbalance -- the global class imbalance degree (MID) and the local difference of class imbalance among clients (WCS). Then, we conduct extensive experiments to analyze the impact of class imbalance on the global performance in various scenarios based on our definition. Our results show that a higher MID and a larger WCS degrade more the performance of the global model. Besides, WCS is shown to slow down the convergence of the global model by misdirecting the optimization.
Particle beam eigen-emittances comprise the lowest set of rms-emittances that can be imposed to a beam through symplectic optical elements. For cases of practical relevance this paper introduces an approximation providing a very simple and powerful r
elation between transverse eigen-emittance variation and the beam phase integral. This relation enormously facilitates modeling eigen-emittance tailoring scenarios. It reveals that difference of eigen-emittances is given by the beam phase integral or vorticity rather than by angular momentum. Within the approximation any beam is equivalent to two objects rotating at angular velocities of same strength and different sign. A description through circular beam modes has been done already in [A. Burov, S. Nagaitsev, and Y. Derbenev, Circular modes, beam adapters, and their applications in beam optics, Phys. Rev. E 66, 016503 (2002)]. The new relation presented here is a complementary and vivid approach to provide a physical picture of the nature of eigen-emittances for cases of practical interest.
The study is devoted to search for flare stars among confirmed members of Galactic open clusters using high-cadence photometry from {it TESS} mission. We analyzed 957 high-cadence light curves of members from 136 open clusters. As a result, 56 flare
stars were found, among them 8 hot B-A type objects. Of all flares, 63% were detected in a sample of cool stars ($T_{rm eff}<5000$~K), and 29% -- in stars of spectral type G, while 23% in K-type stars and approximately 34% of all detected flares are in M-type stars. Using the FLATWRM (FLAre deTection With Ransac Method) flare finding algorithm, we estimated parameters of flares and rotation period of detected flare stars. The flare with the largest amplitude appears on the M3 type EQ,Cha star. Statistical analysis did not reveal any direct correlation between ages, rotation periods and flaring activity.
A dedicated device to fully determine the four-dimensional beam matrix, called ROSE (ROtating System for Emittance measurements) was successfully commissioned. Results obtained with 83Kr13+ at 1.4 MeV/u are reported in Phys. Rev. Accel. Beams 19, 072
802 (2016). Coupled moments were determined with an accuracy of about 10%, which is sufficiently low to reliably determine a lattice which could decouple the beam. However, the remaining uncertainty on the corresponding eigen emittances was still considerable high. The present paper reports on improvement of the evaluation procedure which lowers the inaccuracy of measured eigen emittances significantly to the percent level. The method is based on trimming directly measured data within their intrinsic measurement resolution such that the finally resulting quantity is determined with high precision.
Coagulation of dust aggregates plays an important role in the formation of planets and is of key importance to the evolution of protoplanetary disks (PPDs). Characteristics of dust, such as the diversity of particle size, porosity, charge, and the ma
nner in which dust couples to turbulent gas, affect the collision outcome and the rate of dust growth. Here we present a numerical model of the evolution of the dust population within a PPD which incorporates all of these effects. The probability that any two particles collide depends on the particle charge, cross-sectional area and their relative velocity. The actual collision outcome is determined by a detailed collision model which takes into account the aggregate morphology, trajectory, orientation, and electrostatic forces acting between charged grains. The data obtained in this research reveal the characteristics of dust populations in different environments at the end of the hit-and-stick growth, which establishes the foundation for the onset of the next growth stage where bouncing, mass transfer and fragmentation become important. For a given level of turbulence, neutral and weakly charged particles collide more frequently and grow faster than highly charged particles. However, highly charged particles grow to a larger size before reaching the bouncing barrier, and exhibit a Runaway growth, in which a few large particles grow quickly by accreting smaller particles while the rest of the population grows very slowly. In general, highly charged aggregates have a more compact structure and are comprised of larger monomers than neutral/weakly charged aggregates. The differences in the particle structure/composition not only affect the threshold velocities for bouncing and fragmentation, but also change the scattering and absorption opacity of dust, influencing the appearance of PPDs.
The long-term evolution of the centroid energy of the CRSF in Her X-1 is still a mystery. We report a new measurement from a campaign between {sl Insight}-HXMT and {sl NuSTAR} performed in February 2018. Generally, the two satellites show well consis
tent results of timing and spectral properties. The joint spectral analysis confirms that the previously observed long decay phase has ended, and that the line energy instead keeps constant around 37.5 keV after flux correction.
In order to characterize the early growth of fine-grained dust rims (FGRs) that commonly surround chondrules, we perform numerical simulations of dust accretion onto chondrule surfaces. We employ a Monte Carlo algorithm to simulate the collision of d
ust monomers having radii between 0.5 and 10 $mu$m with chondrules whose radii are between 500 and 1000 $mu$m, in 100-$mu$m increments. The collisions are driven by Brownian motion and solar nebula turbulence. After each collision, the colliding particles either stick at the point of contact, roll or bounce. We limit accretion of dust monomers (and in some cases, dust aggregates) to a small patch of the chondrule surface, for computational expediency. We model the morphology of the dust rim and the trajectory of the dust particle, which are not considered in most of the previous works. Radial profiles of FGR porosity show that rims formed in weak turbulence are more porous (with a porosity of 60-74%) than rims formed in stronger turbulence (with a porosity of 52-60%). The lower end of each range corresponds to large chondrules and the upper end to small chondrules, meaning that the chondrule size also has an impact on FGR porosity. The thickness of FGRs depends linearly on chondrule radius, and the slope of this linear dependency increases with time, and decreases with the turbulence strength. The porosity of FGRs formed by dust aggregates is $sim 20%$ on average greater than that of FGRs formed by single monomers. In general, the relatively high porosities that we obtain are consistent with those calculated by previous authors from numerical simulations, as well as with initial FGR porosities inferred from laboratory measurements of rimmed chondrule samples and rimmed chondrule analogs.
Solar prominences are subject to all kinds of perturbations during their lifetime, and frequently demonstrate oscillations. The study of prominence oscillations provides an alternative way to investigate their internal magnetic and thermal structures
as the oscillation characteristics depend on their interplay with the solar corona. Prominence oscillations can be classified into longitudinal and transverse types. We perform three-dimensional ideal magnetohydrodynamic simulations of prominence oscillations along a magnetic flux rope, with the aim to compare the oscillation periods with those predicted by various simplified models and to examine the restoring force. We find that the longitudinal oscillation has a period of about 49 minutes, which is in accordance with the pendulum model where the field-ligned component of gravity serves as the restoring force. In contrast, the horizontal transverse oscillation has a period of about 10 minutes and the vertical transverse oscillation has a period of about 14 minutes, and both of them can be nicely fitted with a two-dimensional slab model. We also find that the magnetic tension force dominates most of the time in transverse oscillations, except for the first minute when magnetic pressure overwhelms.
In 1926, H. Busch formulated a theorem for one single charged particle moving along a region with a longitudinal magnetic field [H. Busch, Berechnung der Bahn von Kathodenstrahlen in axial symmetrischen electromagnetischen Felde, Z. Phys. 81 (5) p. 9
74, (1926)]. The theorem relates particle angular momentum to the amount of field lines being enclosed by the particle cyclotron motion. This paper extends the theorem to many particles forming a beam without cylindrical symmetry. A quantity being preserved is derived, which represents the sum of difference of eigen-emittances, magnetic flux through the beam area, and beam rms-vorticity multiplied by the magnetic flux. Tracking simulations and analytical calculations using the generalized Courant-Snyder formalism confirm the validity of the extended theorem. The new theorem has been applied for fast modelling of experiments with electron and ion beams on transverse emittance re-partitioning conducted at FERMILAB and at GSI.
Superconductivity of transition metal dichalcogenide $1T$-TiTe$_2$ under high pressure was investigated by the first-principles calculations. Our results show that the superconductivity of $1T$-TiTe$_2$ exhibits very different behavior under the hydr
ostatic and uniaxial pressure. The hydrostatic pressure is harmful to the superconductivity, while the uniaxial pressure is beneficial to the superconductivity. Superconducting transition temperature $T_C$ at ambient pressure is 0.73 K, and it reduces monotonously under the hydrostatic pressure to 0.32 K at 30 GPa. While the $T_C$ increases dramatically under the uniaxial pressure along $c$ axis. The established $T_C$ of 6.34 K under the uniaxial pressure of 17 GPa, below which the structural stability maintains, is above the liquid helium temperature of 4.2 K. The increase of density of states at Fermi level, the redshift of $F(omega)$/$alpha^2F(omega)$ and the softening of the acoustic modes with pressure are considered as the main reasons that lead to the enhanced superconductivity under uniaxial pressure. In view of the previously predicted topological phase transitions of $1T$-TiTe$_2$ under the uniaxial pressure [Phys. Rev. B 88, 155317 (2013)], we consider $1T$-TiTe$_2$ as a possible candidate in transition metal chalcogenides for exploring topological superconductivity.
