Explore the vast range of titles available.

Resistive switching, a phenomenon in which the resistance of a device can be modified by applying an electric field 1 – 5 , is at the core of emerging technologies such as neuromorphic computing and resistive memories 6 – 9 . Among the different types of resistive switching, threshold firing 10 – 14 is one of the most promising, as it may enable the implementation of artificial spiking neurons 7 , 13 , 14 . Threshold firing is observed in Mott insulators featuring an insulator-to-metal transition 15 , 16 , which can be triggered by applying an external voltage: the material becomes conducting (‘fires’) if a threshold voltage is exceeded 7 , 10 – 12 . The dynamics of this induced transition have been thoroughly studied, and its underlying mechanism and characteristic time are well documented 10 , 12 , 17 , 18 . By contrast, there is little knowledge regarding the opposite transition: the process by which the system returns to the insulating state after the voltage is removed. Here we show that Mott nanodevices retain a memory of previous resistive switching events long after the insulating resistance has recovered. We demonstrate that, although the device returns to its insulating state within 50 to 150 nanoseconds, it is possible to re-trigger the insulator-to-metal transition by using subthreshold voltages for a much longer time (up to several milliseconds). We find that the intrinsic metastability of first-order phase transitions is the origin of this phenomenon, and so it is potentially present in all Mott systems. This effect constitutes a new type of volatile memory in Mott-based devices, with potential applications in resistive memories, solid-state frequency discriminators and neuromorphic circuits. Mott materials feature scale-less relaxation dynamics after the insulator-to-metal transition that make its electric triggering dependent on recent switching events.

Instantaneous frequency measurement devices are designated for very fast measurements of the current frequency value of microwave signals, even if they are very short in the time domain. Fast measurements of frequency temporary values may be based on the evaluation of the phase difference of signal propagating through the microwave transmission lines with unequal, but known, lengths. This paper presents the principle of determination of temporary values of the microwave signal frequency using the digitalized signals and the binary value of them eventually. In the purpose of increase the frequency discrimination resolution, additional tracks with lines with a larger length are proposed. For the system with elements with analytical model transmission characteristics it is typical that bands of ambiguity of frequency measurement occurs. To tackle this problem in addition to 4 x 4 Butler matrix implementation the method of using combination sine and cosine signals is proposed.

The orthogonal frequency discrimination is achieved by utilizing waveform transformation in the phase-shift multiplier frequency discriminator. Firstly, the constraint relationship is derived between the circuit parameters of the phase shifter and the maximum frequency spectrum of the input FM signal. Subsequently, the demodulation system for the input FM signal is designed using a combination of a phase shifter and multiplier. Finally, the maximum frequency offset range of distortion-free output from the frequency discriminator is analyzed, and the influence of resistance and capacitance parameters in the phase-shift network on the input frequency offset range is researched using parameter scanning analysis. Theoretical derivation analysis and simulation testing confirm that this designed circuit can effectively demodulate FM signals with a wide range of non-distortion when the circuit parameters match those of input signals. This circuit simulation method has been proven to visually demonstrate different output waveforms under various parameter settings, thereby facilitating the visualization of abstract theoretical knowledge.

Discriminator Guidance has become a popular method for efficiently refining pre-trained Score-Matching Diffusion models. However, in this paper, we demonstrate that the standard implementation of this technique does not necessarily lead to a distribution closer to the real data distribution. Specifically, we show that training the discriminator using Cross-Entropy loss, as commonly done, can in fact increase the Kullback-Leibler divergence between the model and target distributions, particularly when the discriminator overfits. To address this, we propose a theoretically sound training objective for discriminator guidance that properly minimizes the KL divergence. We analyze its properties and demonstrate empirically across multiple datasets that our proposed method consistently improves over the conventional method by producing samples of higher quality.

Hyperspectral image (HSI) classification across varying geographical scenes demonstrates significant potential in mitigating data discrepancies caused by environmental diversity, particularly in operational domains requiring robust remote sensing interpretation. Current several approaches exhibit improved generalization performance in the domain of cross-scene hyperspectral image classification. However, object similarity, data heterogeneity, and imbalanced sample distribution remain critical challenges that significantly constrain the generalization capability of cross-scene hyperspectral classification. To address these problems, we propose a Causal Inference inspired Network for hyperspectral image cross-scene classification (CINet). The Causal Inference Block based on the Random Fourier Features is introduced in the discriminator, which eliminates spurious feature correlations and enhances focus on relevant feature-label relationships by sample weighting. To enhance the robustness and generalization of the cross-scene task, we incorporate supervised contrast learning and contrast-adversarial training in the discriminator, where the former optimizes the sample distance and the latter suppresses inter-domain differences by generating extended domains with significant differences. Finally, we conduct extensive experiments on two public datasets: Houston and Pavia. Compared with two Domain Adaptation methods and three Domain Generalization methods, the overall accuracy of our proposed CINet is more than 5% higher than the two DA methods, and 0.5%-2.5% higher than the three DG methods. The experimental results highlight the superior performance of CINet, confirming its effectiveness in cross-scene classification task.

Deep learning-based computer vision technology significantly improves the accuracy and efficiency of crop disease detection. However, the scarcity of crop disease images leads to insufficient training data, limiting the accuracy of disease recognition and the generalization ability of deep learning models. Therefore, increasing the number and diversity of high-quality disease images is crucial for enhancing disease monitoring performance. We design a frequency-domain and wavelet image augmentation network with a dual discriminator structure (FHWD). The first discriminator distinguishes between real and generated images, while the second high-frequency discriminator is specifically used to distinguish between the high-frequency components of both. High-frequency details play a crucial role in the sharpness, texture, and fine-grained structures of an image, which are essential for realistic image generation. During training, we combine the proposed wavelet loss and Fast Fourier Transform loss functions. These loss functions guide the model to focus on image details through multi-band constraints and frequency domain transformation, improving the authenticity of lesions and textures, thereby enhancing the visual quality of the generated images. We compare the generation performance of different models on ten crop diseases from the PlantVillage dataset. The experimental results show that the images generated by FHWD contain more realistic leaf disease lesions, with higher image quality that better aligns with human visual perception. Additionally, in classification tasks involving nine types of tomato leaf diseases from the PlantVillage dataset, FHWD-enhanced data improve classification accuracy by an average of 7.25% for VGG16, GoogleNet, and ResNet18 models.Our results show that FHWD is an effective image augmentation tool that effectively addresses the scarcity of crop disease images and provides more diverse and enriched training data for disease recognition models.

According to the requirement of microwave frequency measurement, a microwave frequency measurement system is designed based on six-port device. The system includes a microwave power divider, delay lines, a microwave phase discriminator, detectors, a signal source, a oscilloscope and so on. The microwave phase discriminator is realized by six-port circuit. The determination method of delay line length is explained, the phase calculation method of six–port phase discriminator is deduced, and the relationship between frequency, phase difference and delay line is given. According to the method, a microwave frequency measurement system is built to measure the sine wave signal with frequency of 2GHz to 8GHz. The best relative error of the measurement is 1.9×10 −3 . The results show that the six-port device can be applied to the microwave frequency measurement system. In the future, the research on wide-band and high accuracy frequency measurement technology will be further carried out.

Speech-driven facial animation is the process that automatically synthesizes talking characters based on speech signals. The majority of work in this domain creates a mapping from audio features to visual features. This approach often requires post-processing using computer graphics techniques to produce realistic albeit subject dependent results. We present an end-to-end system that generates videos of a talking head, using only a still image of a person and an audio clip containing speech, without relying on handcrafted intermediate features. Our method generates videos which have (a) lip movements that are in sync with the audio and (b) natural facial expressions such as blinks and eyebrow movements. Our temporal GAN uses 3 discriminators focused on achieving detailed frames, audio-visual synchronization, and realistic expressions. We quantify the contribution of each component in our model using an ablation study and we provide insights into the latent representation of the model. The generated videos are evaluated based on sharpness, reconstruction quality, lip-reading accuracy, synchronization as well as their ability to generate natural blinks.

A bstract We devise an autoencoder based strategy to facilitate anomaly detection for boosted jets, employing Graph Neural Networks (GNNs) to do so. To overcome known limitations of GNN autoencoders, we design a symmetric decoder capable of simultaneously reconstructing edge features and node features. Focusing on latent space based discriminators, we find that such setups provide a promising avenue to isolate new physics and competing SM signatures from sensitivity-limiting QCD jet contributions. We demonstrate the flexibility and broad applicability of this approach using examples of W bosons, top quarks, and exotic hadronically-decaying exotic scalar bosons.

Imitation learning aims to solve the problem of defining reward functions in real-world decision-making tasks. The current popular approach is the Adversarial Imitation Learning (AIL) framework, which matches expert state-action occupancy measures to obtain a surrogate reward for forward reinforcement learning. However, the traditional discriminator is a simple binary classifier and doesn't learn an accurate distribution, which may result in failing to identify expert-level state-action pairs induced by the policy interacting with the environment. To address this issue, we propose a method named diffusion adversarial imitation learning (DiffAIL), which introduces the diffusion model into the AIL framework. Specifically, DiffAIL models the state-action pairs as unconditional diffusion models and uses diffusion loss as part of the discriminator's learning objective, which enables the discriminator to capture better expert demonstrations and improve generalization. Experimentally, the results show that our method achieves state-of-the-art performance and significantly surpasses expert demonstration on two benchmark tasks, including the standard state-action setting and state-only settings. Our code can be available at the link https://github.com/ML-Group-SDU/DiffAIL.