In this paper, we designed an ultra-wideband solar energy absorber and approved it numerically by the finite-difference time-domain simulation. The designed solar energy absorber can achieve a high absorption of more than 90% of light in a continuous 3.506 μm (0.596 μm–4.102 μm) wavelength range. The basic structure of the absorber is based on silicon dioxide colloidal crystal and Ti. Since the materials have a high melting point, the designed solar energy absorber can work normally under high temperature, and the structure of this solar energy absorber is simpler than most solar energy absorbers fabricated with traditional metal. In the entire wavelength band researched, the average absorption of the colloidal crystal-based solar energy absorber is as high as 94.3%, demonstrating an excellent performance under the incidence light of AM 1.5 solar spectrum. In the meantime, the absorption spectrum of the solar energy absorber is insensitive to the polarization of light. In comparison to other similar structures, our designed solar energy absorber has various advantages, such as its high absorption in a wide spectrum range and that it is low cost and easy to make.

Deep neural networks are often susceptible to overfitting, necessitating effective regularization techniques. This paper introduces Spectral Adaptive Dropout, a novel frequency-based regularization technique that dynamically adjusts dropout rates based on the spectral characteristics of network gradients. The proposed approach addresses the limitations of traditional dropout methods by adaptively targeting high-frequency components that typically contribute to overfitting while preserving essential low-frequency information. Through extensive experimentation on character-level language modeling tasks, the study demonstrates that the method achieves a 1.10% improvement in validation loss while maintaining competitive inference speeds. Thise research explores several implementations including FFT-based analysis, wavelet decomposition, and per-attention-head adaptation, culminating in an optimized approach that balances computational efficiency with regularization effectiveness. Our results highlight the significant potential of incorporating frequency-domain information into regularization strategies for deep neural networks.

This paper challenges the prevailing assumption that Mixture of Experts (MoE) consistently improves computational efficiency through a systematic evaluation of MoE variants in Transformer models. We implement and compare three approaches: basic MoE, top-k routing, and capacity-factored routing, each progressively addressing load-balancing challenges. Our experiments reveal critical trade-offs between performance and efficiency: while MoE models maintain validation performance comparable to baselines, they require significantly longer training times (a 50% increase) and demonstrate reduced inference speeds (up to 56% slower). Analysis of routing behavior shows that even with load-balancing techniques, expert utilization remains unevenly distributed. These findings provide empirical evidence that MoE’s computational benefits are highly dependent on model scale and task characteristics, challenging common assumptions about sparse architectures and offering crucial guidance for adaptive neural architecture design across different computational constraints.

Multi-band circular dichroism (CD) with spectral tunability is highly desirable for chiral metasurface-based sensing and polarization control. In this work, we propose a mid-infrared (MIR) metal–insulator–metal (MIM) chiral metasurface absorber composed of gold-alumina-gold (Au-Al₂O₃-Au) layers, where chirality is introduced by symmetry breaking between two gold elliptical bars in the top layer. Finite element calculations show that the structure operates over 3.5–6.5 µm and produces dual-band CD responses with values of 0.83 and −0.81. The CD magnitude in each band can be independently tuned by adjusting the semi-major or semi-minor axis of the elliptical bars. In addition, rectangular slots inserted into the bars enable continuous redshift of the resonance wavelengths, and the tunability can be further enhanced by optimizing the slot dimensions. These results provide a practical strategy for designing tunable dual-band chiral absorbers and may be useful for future chiral sensing and polarization imaging applications.