No abstract

In class incremental learning (CIL) a model must learn new classes in a sequential manner without forgetting old ones. However, conventional CIL methods consider a balanced distribution for each new task, which ignores the prevalence of long-tailed distributions in the real world. In this work we propose two long-tailed CIL scenarios, which we term ordered and shuffled LT-CIL. Ordered LT-CIL considers the scenario where we learn from head classes collected with more samples than tail classes which have few. Shuffled LT-CIL, on the other hand, assumes a completely random long-tailed distribution for each task. We systematically evaluate existing methods in both LT-CIL scenarios and demonstrate very different behaviors compared to conventional CIL scenarios. Additionally, we propose a two-stage learning baseline with a learnable weight scaling layer for reducing the bias caused by long-tailed distribution in LT-CIL and which in turn also improves the performance of conventional CIL due to the limited exemplars. Our results demonstrate the superior performance (up to 6.44 points in average incremental accuracy) of our approach on CIFAR-100 and ImageNet-Subset. The code is available at https://github.com/xialeiliu/Long-Tailed-CIL.

Microwave heating of solid stack materials is common but bothered by problems of uneven heating and electric discharge phenomena. In this paper, a method introducing fluid materials with different relative permittivity is proposed to improve the heating uniformity and safety of solid stack materials. Simulations have been computed based on the finite element method (FEM) and validated by experiments. Simulation results show that the introducing of fluid materials with proper relative permittivity does improve the heating uniformity and safety. Fluid materials with the larger real part of relative permittivity could obviously lower the maximum modulus value of the electric field for about 23 times, and will lower the coefficient of variation (COV) in general, although in small ranges that it has fluctuated. Fluid materials with the larger imaginary part of relative permittivity, in a range from 0 to 0.3, can make a more efficient heating and it could lower the maximum modulus value of the electric field by 34 to 55% on the whole studied range. However, the larger imaginary part of relative permittivity will cause worse heating uniformity as the COV rises by 246.9% in the same process. The computed results are discussed and methods to reach uniform and safe heating through introducing fluid materials with proper relative permittivity are proposed.

The local overheating may occur in microwave heating process, which results in poor heating uniformity. In order to improve the heating performance, in this paper, the boundary movement with periodical movement is added to the traditional rotating turntable model. The continuous algorithm based on moving mesh is adopted to simulate the heating process, and the results of calculation show that the proposed method can not only maintain the uniformity of the traditional rotating turntable method but also increase nearly 30% of the average temperature of the heated object.

The response of uniaxial anisotropic ferromagnetic particles with linear reaction dynamics subjected to alternating current (AC) or direct current (DC) bias magnetic field is evaluated by the reaction–diffusion equation for the probability distribution function of the molecular concentration in the spherical coordinate system. The magnetization function and the probability distribution function of the magnetic particles in the reaction system are derived by using the Legendre polynomials and Laplace transform. We discuss the characteristics of magnetization and probability distribution of the magnetic particles with different anisotropic parameters driven by a DC and AC magnetic fields, respectively. It is shown that both the magnetization and the probability distribution decrease with time increasing due to the reaction process. The uniformity of the probability distribution and the amplitude of the magnetization are both affected by the anisotropic parameters. Meanwhile, the difference between the case with linear reaction dynamics and the non-reaction case is discussed.