Large language models (LLMs) primarily trained on English texts, often face biases and inaccuracies in Chinese contexts. Their limitations are pronounced fields like Traditional Medicine (TCM), where cultural clinical subtleties vital, further hindered by a lack of domain-specific data, such as rheumatoid arthritis (RA). To address these issues, this paper introduces Hengqin-RA-v1, the first large model specifically tailored for TCM with focus diagnosing treating RA. We also present...

ABSTRACT We have investigated a number of factors that can significant impacts on the classification performance gamma-ray sources detected by Fermi Large Area Telescope (LAT) with machine learning techniques. show framework automatic feature selection construct simple model small set features yields better over previous results. Secondly, because sample size training/test sets certain classes in gamma-ray, nested re-sampling and cross-validations are suggested for quantifying statistical...

Token interaction operation is one of the core modules in MLP-based models to exchange and aggregate information between different spatial locations. However, power token on dimension highly dependent resolution feature maps, which limits model's expressive ability, especially deep layers where are down-sampled a small size. To address this issue, we present novel method called Strip-MLP enrich three ways. Firstly, introduce new MLP paradigm Strip layer that allows interact with other tokens...

Multimodal learning with incomplete input data (missing modality) is very practical and challenging. In this work, we conduct an in-depth analysis of challenge find that modality dominance has a significant negative impact on the model training, greatly degrading missing performance. Motivated by Grad-CAM, introduce novel indicator, gradients, to monitor reduce which widely exists in missing-modality scenario. aid present Gradient-guided Modality Decoupling (GMD) method decouple dependency...

We report the results of searching pulsar-like candidates from unidentified objects in $3^{\rm rd}$ Catalog Hard Fermi-LAT sources (3FHL). Using a machine-learning based classification scheme with nominal accuracy $\sim98\%$, we have selected 27 200 3FHL for an identification campaign. archival data, X-ray are found within $\gamma-$ray error ellipses 10 candidates. Within circles much better constrained positions, also searched optical/infrared counterparts and examined their spectral energy...

To monitor road safety, billions of records can be generated by Controller Area Network bus each day on public transportation. Automation to determine whether certain driving behaviour drivers transportation considered safe the using artificial intelligence or machine learning techniques for big data analytics has become a possibility recently. Due high false classification rates current methods, our goal is build practical and accurate method safety predictions that automatically if In this...

With fast growing data collected by the Fermi Large Area Telescope as a big problem, manual classification has become an impossible task for astronomers. In this paper, we propose novel framework using machine learning techniques gamma-ray object classification. We use random forest (RF) algorithm feature selection in order to achieve better performance. After extensive experimental study with incorporated, found best results can be obtained both active galactic nuclei (AGN) / pulsars (PSR)...

In conventional few-shot learning approaches, masked image modeling paradigms such as autoencoders are typically used feature extractors, followed by classifiers. Traditional depend on standard transformers for encoding, neglecting the inherent inductive biases of CNNs, which pivotal efficiency. response, this paper introduces Modernized Convolutional Network with Efficient Channel Attention (ConvNeXt-ECA), a effective encoder that modernizes vanilla ResNet through integration...

Multimodal learning with incomplete input data (missing modality) is practical and challenging. In this work, we conduct an in-depth analysis of challenge find that modality dominance has a significant negative impact on the model training, greatly degrading missing performance. Motivated by Grad-CAM, introduce novel indicator, gradients, to monitor reduce which widely exists in missing-modality scenario. aid present Gradient-guided Modality Decoupling (GMD) method decouple dependency...

Abstract Facial expression is a key to nonverbal communication, which has been confirmed by many different research projects. A change in intensity or magnitude of even one specific facial can cause interpretations. With the continuous and fast development computer vision pattern recognition, recognition received significant attention recently due wide range commercial law enforcement application availability feasible technology during 30 years research. In this thesis, studied applying...

Radiomic representations can quantify properties of regions interest in medical image data. Classically, they account for pre-defined statistics shape, texture, and other low-level features. Alternatively, deep learning-based are derived from supervised learning but require expensive annotations experts often suffer overfitting data imbalance issues. In this work, we address the challenge 3D images an effective quantification under imbalance. We propose a \emph{self-supervised}...

Token interaction operation is one of the core modules in MLP-based models to exchange and aggregate information between different spatial locations. However, power token on dimension highly dependent resolution feature maps, which limits model's expressive ability, especially deep layers where are down-sampled a small size. To address this issue, we present novel method called \textbf{Strip-MLP} enrich three ways. Firstly, introduce new MLP paradigm Strip layer that allows interact with...

In this paper, a novel approach to identifying superpixels in the cluttered environment is proposed. our proposed method, temporal cue and depth maps obtained from sensors are combined with popular method SLIC for using new formulation of distance-minimizing clustering. Under environment, can, compared color-based approaches, better identify contour objects. Experiments have been carried out public dataset compare other methods. The experimental results demonstrate that outperforms approaches.

The new aggregate mode has appeared in China's car-hailing market. allows separate firms to join one single platform and share users transportation capacity together, which can promote the compatibility level among accessed platforms. This paper builds a duopoly model of two-sided market analyzes effects on from perspective compatibility. It is considered this that symmetric travelers multi-homing while drivers single-homing. main results indicate at early stage firms, they tend be exclusive...