Symbolic regression (SR) has emerged as a pivotal technique for uncovering the intrinsic information within data and enhancing the interpretability of AI models. However, current state-of-the-art (sota) SR methods struggle to perform correct recovery of symbolic expressions from high-noise data. To address this issue, we introduce a novel noise-resilient SR (NRSR) method capable of recovering expressions from high-noise data. Our method leverages a novel reinforcement learning (RL) approach in conjunction with a designed noise-resilient gating module (NGM) to learn symbolic selection policies. The gating module can dynamically filter the meaningless information from high-noise data, thereby demonstrating a high noise-resilient capability for the SR process. And we also design a mixed path entropy (MPE) bonus term in the RL process to increase the exploration capabilities of the policy. Experimental results demonstrate that our method significantly outperforms several popular baselines on benchmarks with high-noise data. Furthermore, our method also can achieve sota performance on benchmarks with clean data, showcasing its robustness and efficacy in SR tasks.

Deep reinforcement learning (DRL) has emerged as a leading technique for designing AI-bots in the gaming industry. However, practical implementation of DRL-trained bots often encounter two significant challenges: improving strength and diversifying strategies to satisfy player expectations. We observe that the strength of AI-bots are intrinsically tied to the diversity of emerged strategies. Considering this relationship, we introduce diversity is strength (DIS), a novel DRL training framework capable of concurrently training multiple types of AI-bots for adversarial games. These bots are interconnected through an elaborated history model pool (HMP) structure, thereby improving their strength and strategy diversity to tackle the aforementioned challenges. We further devise a model evaluation and sampling scheme to form the HMP, identify superior models, and enrich the model strategies. The DIS can generate diverse and reliable strategies without the need for human data. This method is validated by achieving first-place finishes in two AI competitions based on complex adversarial games, including Google Research Football and Olympic Games. Experiments demonstrate that bots trained using DIS attain an excellent performance and plentiful strategies. Specifically, diversity analysis demonstrates that the trained bots possess a wealth of strategies, and ablation studies confirm the beneficial impact of the designed modules on the training process.

Deep learning-based diagnosis is becoming an indispensable part of modern healthcare. For high-performance diagnosis, the optimal design of deep neural networks (DNNs) is a prerequisite. Despite its success in image analysis, existing supervised DNNs based on convolutional layers often suffer from their rudimentary feature exploration ability caused by the limited receptive field and biased feature extraction of conventional convolutional neural networks (CNNs), which compromises the network performance. Here, we propose a novel feature exploration network named manifold embedded multilayer perceptron (MLP) mixer (ME-Mixer), which utilizes both supervised and unsupervised features for disease diagnosis. In the proposed approach, a manifold embedding network is employed to extract class-discriminative features; then, two MLP-Mixer-based feature projectors are adopted to encode the extracted features with the global reception field. Our ME-Mixer network is quite general and can be added as a plugin to any existing CNN. Comprehensive evaluations on two medical datasets are performed. The results demonstrate that their approach greatly enhances the classification accuracy in comparison with different configurations of DNNs with acceptable computational complexity.

Embedding high-dimensional data onto a low-dimensional manifold is of both theoretical and practical value. In this article, we propose to combine deep neural networks (DNN) with mathematics-guided embedding rules for high-dimensional data embedding. We introduce a generic deep embedding network (DEN) framework, which is able to learn a parametric mapping from high-dimensional space to low-dimensional space, guided by well-established objectives such as Kullback-Leibler (KL) divergence minimization. We further propose a recursive strategy, called deep recursive embedding (DRE), to make use of the latent data representations for boosted embedding performance. We exemplify the flexibility of DRE by different architectures and loss functions, and benchmarked our method against the two most popular embedding methods, namely, t-distributed stochastic neighbor embedding (t-SNE) and uniform manifold approximation and projection (UMAP). The proposed DRE method can map out-of-sample data and scale to extremely large datasets. Experiments on a range of public datasets demonstrated improved embedding performance in terms of local and global structure preservation, compared with other state-of-the-art embedding methods. Code is available at https://github.com/tao-aimi/DeepRecursiveEmbedding.

Recently, the use of portable equipment has attracted much attention in the medical ultrasound field. Handheld ultrasound devices have great potential for improving the convenience of diagnosis, but noise-induced artifacts and low resolution limit their application. To enhance the video quality of handheld ultrasound devices, we propose a low-rank representation multipathway generative adversarial network (LRR MPGAN) with a cascade training strategy. This method can directly generate sequential, high-quality ultrasound video with clear tissue structures and details. In the cascade training process, the network is first trained with plane wave (PW) single-/multiangle video pairs to capture dynamic information and then fine-tuned with handheld/high-end image pairs to extract high-quality single-frame information. In the proposed GAN structure, a multipathway generator is applied to implement the cascade training strategy, which can simultaneously extract dynamic information and synthesize multiframe features. The LRR decomposition channel approach guarantees the fine reconstruction of both global features and local details. In addition, a novel ultrasound loss is added to the conventional mean square error (MSE) loss to acquire ultrasound-specific perceptual features. A comprehensive evaluation is conducted in the experiments, and the results confirm that the proposed method can effectively reconstruct high-quality ultrasound videos for handheld devices. With the aid of the proposed method, handheld ultrasound devices can be used to obtain convincing and convenient diagnoses.

In the medical ultrasound field, ultrafast imaging has recently become a hot topic. However, the diagnostic reliability of ultrafast high-frame rate plane-wave (PW) imaging is reduced by its low-quality images. The medical ultrasound equipment on the market usually adopts the line-scanning mode, which can obtain high-quality images at a very low frame rate. In addition, many proven data-driven ultrasound image processing methods are trained by line-scan images. Since the gray-level distributions of line-scan images and PW images are very different, these gray-level distribution-sensitive methods cannot be generalized to ultrafast ultrasound imaging, which limits further applications. Hence, we propose an ultrasound-transfer generative adversarial network to improve the quality of PW images and extend the existing image processing methods to ultrafast ultrasound imaging by reconstructing PW images into line-scan images. This network adopts a residual dense generator with a self-attention system that fully uses the hierarchical features and generates details from all the relevant physiological information. A projection discriminator and spectral normalization are introduced to increase the discernibility and to maintain a balance between the generator and the discriminator. Moreover, we reorganize the transmit sequence of the transducer array to eliminate the negative influence of human movements and facilitate the convergence of the proposed model. The experimental results are evaluated with five metrics, which confirm the feasibility of the proposed method to obtain a line-scan-quality image with a very high frame rate. This technology could significantly popularize ultrafast medical ultrasound imaging.