Science China Technological Sciences最新文献

Paper-based electronics offer a simple and cost-effective means to fabricate reconfigurable devices. In response to the problem of fixed shape and single function of most antennas, which limits their applications, a reconfigurable paper-based metamaterial antenna with 2D and 3D forms is presented for tunable operating frequency. The proposed antenna consists of two square split resonant rings fed by a coplanar waveguide. The working frequency of the 2D antenna is adjusted by the length, width, and opening size of the internal open resonant ring. While the folding angle of the antenna turns from 0° to 90°, the operating frequency of the paper-based metamaterial antenna changes from 2.759 to 4.223 GHz. In terms of 3D form, an additional resonant peak is generated by bending the paper-based metamaterial antenna, thus further realizing dual-band antenna design. After a simple process flow, a series of proposed antennas are fabricated and evaluated. The simulated and measured results both demonstrate that the proposed antenna has a good performance in turning the working band. The environment-friendly nature and pliability of paper, as well as simple fabrication procedures, make paper-based metamaterial promising candidates for future green electronics.

Efforts are underway to establish a permanent lunar base on the Moon. In situ lunar regolith is anticipated to be useful as a building material after sintering. However, sintering lunar regolith into a large-scale structure presents challenges. Therefore, the key to lunar construction lies in assembling multiple small-sized sintered modules into a stable, large-sized structure. This study explored the feasibility of welding the sintered HUST-1 lunar regolith simulant (HLRS) using a laser device and conducted experiments using lasers of varying power. The microstructure, mineral composition, element distribution, and shear strength of the welded joint were investigated. A few low-melting minerals were fused and vaporized during welding, leading to the generation of thermal decomposition gas. Furthermore, the welded joint exhibited numerous micro-cracks, pores, and bubbles, resulting in reduced weld shear strength. Finally, the influence of laser power on weld shear strength was investigated, revealing that the highest shear strength (15.69 N/cm) was achieved at a laser power of 1000 W. This study demonstrates the feasibility of laser welding of sintered HLRS for the first time, with potential applications in lunar base construction.

Deep neural networks play an important role in the recognition of waste electrical appliances. However, deep neural network components still lack reliability in decision-making features. To address this problem, a sparse convolutional model with semantic expression (SCMSE) is proposed. First, a low-rank sparse semantic expression component, combining the benefits of residual networks and sparse representation, is adapted to enhance sparse feature extraction and semantic expression. Second, a reliable network architecture is obtained by iterating the optimal sparse solution, enhancing semantic expression. Finally, the results of visualization experiments on the waste electrical appliances dataset demonstrate that the proposed SCMSE can obtain excellent semantic performance.

In recent years, the authors have extended the traditional interval method into the time dimension to develop a new mathematical tool called the “interval process model” for quantifying time-varying or dynamic uncertainties. This model employs upper and lower bounds instead of precise probability distributions to quantify uncertainty in a parameter at any given time point. It is anticipated to complement the conventional stochastic process model in the coming years owing to its relatively low dependence on experimental samples and ease of understanding for engineers. Building on our previous work, this paper proposes a spectrum analysis method to describe the frequency domain characteristics of an interval process, further strengthening the theoretical foundation of the interval process model and enhancing its applicability for complex engineering problems. In this approach, we first define the zero midpoint function interval process and its auto/cross-power spectral density (PSD) functions. We also deduce the relationship between the auto-PSD function and the auto-covariance function of the stationary zero midpoint function interval process. Next, the auto/cross-PSD function matrices of a general interval process are defined, followed by the introduction of the concepts of PSD function matrix and cross-PSD function matrix for interval process vectors. The spectrum analysis method is then applied to random vibration problems, leading to the creation of a spectrum-analysis-based interval vibration analysis method that determines the PSD function for the system displacement response under stationary interval process excitations. Finally, the effectiveness of the formulated spectrum-analysis-based interval vibration analysis approach is verified through two numerical examples.

The visual system continuously adapts to the statistical properties of the environment. Existing evidence shows a close resemblance between deep convolutional neural networks (CNNs) and primate visual stream in neural selectivity to naturalistic textures above the primary visual processing stage. This study delves into the mechanisms of perceptual learning in CNNs, focusing on how they assimilate the high-order statistics of natural textures. Our results show that a CNN model achieves a similar performance improvement as humans, as manifested in the learning pattern across different types of high-order image statistics. While L2 was the first stage exhibiting texture selectivity, we found that stages beyond L2 were critically involved in learning. The significant contribution of L4 to learning was manifested both in the modulations of texture-selective responses and in the consequences of training with frozen connection weights. Our findings highlight learning-dependent plasticity in the mid-to-high-level areas of the visual hierarchy. This research introduces an AI-inspired approach for studying learning-induced cortical plasticity, utilizing DCNNs as an experimental framework to formulate testable predictions for empirical brain studies.

Dendrite growth is one of the main challenges in maintaining the service life of all-solid-state lithium-ion batteries. Mechanical stress has been reported to significantly affect dendrite growth. In this study, to explain the effect of mechanical stress on electrochemical reactions in all-solid-state batteries, a modified phase-field model for dendrite growth is proposed by considering the stress-dependent overpotential. Dendrite growth under different mechanical loadings in an all-solid-state battery is investigated using the proposed model. Consistent with previous experimental results, the current result shows that compressive stress inhibits dendrite growth. Considering the stress concentration at the tips of processing-induced microcracks, the effects of the number and distribution of microcracks on dendrite growth are investigated. The results show that the stress-concentration field induced at the tips of cracks or voids can change the morphology of dendrites and decrease their growth rates. This study provides a new perspective for explaining Li dendrite growth under mechanical stress and offers inspiration for prolonging the service life of all-solid-state batteries based on defect and stress regulation, which may be further realized in experiments by filling solid electrolytes with different types of nanofillers.

Artificial intelligence (AI) systems surpass certain human intelligence abilities in a statistical sense as a whole, but are not yet the true realization of these human intelligence abilities and behaviors. There are differences, and even contradictions, between the cognition and behavior of AI systems and humans. With the goal of achieving general AI, this study contains a review of the role of cognitive science in inspiring the development of the three mainstream academic branches of AI based on the three-layer framework proposed by David Marr, and the limitations of the current development of AI are explored and analyzed. The differences and inconsistencies between the cognition mechanisms of the human brain and the computation mechanisms of AI systems are analyzed. They are found to be the cause of the differences and contradictions between the cognition and behavior of AI systems and humans. Additionally, eight important research directions and their scientific issues that need to focus on brain-inspired AI research are proposed: highly imitated bionic information processing, a large-scale deep learning model that balances structure and function, multi-granularity joint problem solving bidirectionally driven by data and knowledge, AI models that simulate specific brain structures, a collaborative processing mechanism with the physical separation of perceptual processing and interpretive analysis, embodied intelligence that integrates the brain cognitive mechanism and AI computation mechanisms, intelligence simulation from individual intelligence to group intelligence (social intelligence), and AI-assisted brain cognitive intelligence.

The primate brain and artificial intelligence (AI) can both be conceptualized as information processing systems, each with its own distinct biological and computational architectures. While there are parallels between them, their respective structural and functional connections show significant differences. In this paper, we examine the central role of feedback processing in both the primate brain and AI systems, which has been shown to be crucial in shaping neural processing. By reviewing the key features of feedback processes in the primate brain, which allows the brain to incorporate prior knowledge, contextual information, and task-demands into early-stage processing, we highlight the divergence in goals and functions between biological and AI systems. Understanding these differences is crucial for elucidating the cognitive capabilities of the primate brain and for addressing computational challenges in AI. In advocating “Cognition-Inspired-Computation”, we suggest that integrating insights from feedback processing in the primate brain into AI research will offer potentially significant improvements for the advancement of AI systems.

In this study, we explored the neural mechanism of global topological perception in the human visual system. We showed strong evidence that the retinotectal pathway in the archicortex of the human brain is responsible for global topological perception, and for modulating the local feature processing in the classical ventral visual pathway. Inspired by this recent cognitive discovery, we developed a novel CogNet architecture to emulate the global-local dichotomy of human visual cognitive mechanisms. The thorough experimental results indicate that the proposed CogNet not only significantly improves image classification accuracies but also effectively addresses the texture bias problem observed in baseline CNN models. We have also conducted mathematical analysis for the generalization gap for general neural networks. Our theoretical derivations suggest that the Hurst parameter, a measure of the curvature of the loss landscape, can closely bind the generalization gap. A larger Hurst parameter corresponds to a better generalization ability. We found that our proposed CogNet achieves a lower test error and attains a larger Hurst parameter, strengthening its superiority over the baseline CNN models further.

The human visual system is a complex and interconnected network comprising billions of neurons. It plays an essential role in translating environmental light stimuli into information that guides and shapes human perception and action. Research on the visual system aims to uncover the underlying neural structure principles of human visual perception and their possible applications. Currently, there are two main approaches: biological system analysis and simulation, artificial intelligence models based on deep learning. Here we aim to discuss the two approaches to human-level vision systems. Deep learning has significantly impacted the field of vision with achievements in representation, modeling, and hardware design. However, there is still a significant gap between deep learning models and the human visual system in terms of scalability, transferability, and sustainability. The progress of the biological visual system can help fill the gap by further understanding the properties and functions of different components of the system. We take the efforts of reconstructing the retina as an example to illustrate that even if we are unable to replicate the visual system on a computer right now, we can still learn a lot by combining existing research outcomes in neuroscience. At the end of the paper, we suggest tracing back to gradually build visual systems from the computational counterpart of biological structures to achieve a human-level vision system in the future.