P-tuning has demonstrated that anchor tokens are beneficial for improving the performance of downstream tasks. However, manual selection of anchor tokens may result in subjective or suboptimal results. In this paper, we present aCat to automatically select anchor tokens. Following the framework of the soft-hard prompt paradigm, aCat achieves automatic prompt template construction. Experiments conducted on natural language understanding benchmarks demonstrate the effectiveness of our proposed method. On the seven SuperGLUE datasets, aCat achieves higher accuracy than P-tuning, with an average accuracy surpassing P-tuning V2.

With 6G communication technology advancing, the demand for Radio Environment Maps (REMs) has increased due to their critical role in network optimization, resource management, and signal coverage in complex environments. However, due to the high cost of sensing, only a small number of discrete radio sampling results can be obtained, which limits their application to specific tasks. To address this problem, we propose a novel method for constructing fine-grained REMs in complex environments based on a generative diffusion model. This method leverages the spatial correlations between sparse data points while incorporating conditional information based on global spatial correlations and geographic relationships. The goal is to construct fine-grained radio environment maps from sparse coarse-grained data. Experimental results demonstrate that our model performs significantly well, achieving an absolute error of only 2.79 dB even when the sampling rate is as low as 10%.

Aquatic information monitoring is crucial for the sustainable management of water environments. Conventional interpolation methods commonly hinge on assumptions of spatial proximity or temporal similarity. However, they often fall short of capturing the intricate spatiotemporal correlations present in water quality sequences, affecting our understanding of the spatial patterns of regional water quality conditions. In this study, we propose a framework for river basin information fine-grained sensing based on deep learning, which includes a global sensing model (SGM) and a static deployment model. Inside the SGM, we adopt a multidimensional convolutional neural network (CNN) to extract spatiotemporal features and an attention mechanism to fuse these features, to infer water quality variable information on unmonitored points. Since the inference outcomes could be affected by the locations of the sensors, to minimize the inference error of the SGM, the static deployment model was designed to aid the deployment of sensors into strategic locations of a river basin to obtain optimum spatial-temporal data samples. The research results not only revealed the spatial distribution patterns of total nitrogen (TN) concentrations but also showed that the proposed method could yield a better inference performance compared to traditional interpolation methods.

Photoacoustic microscopy (PAM) is a promising imaging modality because it is able to reveal optical absorption contrast in high resolution on the order of a micrometer. It can be applied in an endoscopic approach by implementing PAM into a miniature probe, termed photoacoustic endoscopy (PAE). Here we develop a miniature focus-adjustable PAE (FA-PAE) probe characterized by both high resolution (in micrometers) and large depth of focus (DOF) via a novel optomechanical design for focus adjustment. To realize high resolution and large DOF in a miniature probe, a 2-mm plano-convex lens is specially adopted, and the mechanical translation of a single-mode fiber is meticulously designed to allow the use of multi-focus image fusion (MIF) for extended DOF. Compared with existing PAE probes, our FA-PAE probe achieves high resolution of [Formula: see text] within unprecedentedly large DOF of 3.2 mm, more than 27 times the DOF of the probe without performing focus adjustment for MIF. The superior performance is first demonstrated by imaging both phantoms and animals including mice and zebrafish in vivo by linear scanning. Further, in vivo endoscopic imaging of a rat's rectum by rotary scanning of the probe is conducted to showcase the capability of adjustable focus. Our work opens new perspectives for PAE biomedical applications.

Many important problems in science and engineering require solving the so-called parametric partial differential equations (PDEs), i.e., PDEs with different physical parameters, boundary conditions, shapes of computation domains, etc. Recently, building learning-based numerical solvers for parametric PDEs has become an emerging new field. One category of methods such as the Deep Galerkin Method (DGM) and Physics-Informed Neural Networks (PINNs) aim to approximate the solution of the PDEs. They are typically unsupervised and mesh-free, but require going through the time-consuming network training process from scratch for each set of parameters of the PDE. Another category of methods such as Fourier Neural Operator (FNO) and Deep Operator Network (DeepONet) try to approximate the solution mapping directly. Being fast with only one forward inference for each PDE parameter without retraining, they often require a large corpus of paired input-output observations drawn from numerical simulations, and most of them need a predefined mesh as well. In this paper, we propose Meta-Auto-Decoder (MAD), a mesh-free and unsupervised deep learning method that enables the pre-trained model to be quickly adapted to equation instances by implicitly encoding (possibly heterogenous) PDE parameters as latent vectors. The proposed method MAD can be interpreted by manifold learning in infinite-dimensional spaces, granting it a geometric insight. Extensive numerical experiments show that the MAD method exhibits faster convergence speed without losing accuracy than other deep learning-based methods.