The effective integration of artificial intelligence into clinical workflows requires models that go beyond simple prediction to generate comprehensive, explainable, and actionable disease trajectories. Addressing the limitations of opaque deep learning architectures and the noise inherent in electronic health records, we introduce the sequential pattern transformer (SPT), a novel framework that synergizes sequential pattern mining with generative transformer modeling. Using four years of inpatient data from 258,460 type 2 diabetes patients, we applied the PrefixSpan algorithm to distill noisy diagnostic histories into a curated vocabulary of 95,630 statistically validated disease progression patterns. A decoder-only transformer was trained exclusively on these evidence-based sequences to learn the temporal dynamics of disease evolution. This pattern-guided approach shifts the modeling paradigm from classification to probabilistic trajectory generation. The model achieved a robust 85.78% Top-5 accuracy, significantly outperforming a standard LSTM baseline (71.47%). Beyond predictive accuracy, the framework constructs a dynamic Disease Atlas, a branching tree structure that visualizes likely future pathways, augmented by multi-level explainable AI (XAI) including learned clinical clusters, SHAP-based feature attribution, and counterfactual simulations. Crucially, this methodology is domain-agnostic and capable of efficient fine-tuning, making it a transferable solution for adapting to diverse clinical conditions and local hospital settings. SPT thus offers a transparent, robust, and scalable framework for mapping the complex temporal dynamics of disease, bridging the gap between high-performance AI and interpretable clinical application.

Current end-to-end multi-modal models utilize different encoders and decoders to process input and output information. This separation hinders the joint representation learning of various modalities. To unify multi-modal processing, we propose a novel architecture called MDM (Multi-modal Diffusion Mamba). MDM utilizes a Mamba-based multi-step selection diffusion model to progressively generate and refine modality-specific information through a unified variational autoencoder for both encoding and decoding. This innovative approach allows MDM to achieve superior performance when processing high-dimensional data, particularly in generating high-resolution images and extended text sequences simultaneously. Our evaluations in areas such as image generation, image captioning, visual question answering, text comprehension, and reasoning tasks demonstrate that MDM significantly outperforms existing end-to-end models (MonoFormer, LlamaGen, and Chameleon etc.) and competes effectively with SOTA models like GPT-4V, Gemini Pro, and Mistral. Our results validate MDM's effectiveness in unifying multi-modal processes while maintaining computational efficiency, establishing a new direction for end-to-end multi-modal architectures.

Classical acoustic impedance inversion methods rely on mathematical and physical models to estimate subsurface acoustic impedance distribution. However, these methods face difficulties in accurately fitting complex impedance data. While deep learning methods have achieved higher accuracy and efficiency in impedance inversion, they remain black box models, lacking interpretability. So, it is difficult to analyze the reason why they are (or are not) effective. To address the limitations of both traditional and deep learning methods, this paper proposes a novel approach, AII-SR (Acoustic Impedance Inversion with Symbolic Regression), which discovers partial differential equations (PDEs) from impedance data to model acoustic impedance inversion. To discover these PDEs, AII-SR adopts a dual-learning framework. It employs a forward model based on the Robinson convolution principle to ensure the physical consistency and reliability of predictions. Subsequently, AII-SR creates an inversion model that combines a symbolic regression-based PDE generator with a physics-informed solving neural network (PSNN) to identify PDEs that accurately fit the impedance data. Experiments demonstrate that AII-SR outperforms deep learning methods, such as SSEI, TCN and Se-Unet, in terms of accuracy and interpretability. AII-SR generates concise, interpretable mathematical expressions in the form of PDEs, offering profound insights into the physical relationships between seismic and impedance data.

The goal of assembly blocks is to select blocks from pre-trained neural network (NN) models and combine them into a new NN for a different dataset. By reusing the weights of these blocks, training the NN with the new dataset becomes cost-effective. To achieve this goal, we propose an end-to-end differentiable neural network called PA-DNN. PA-DNN consists of two modules: a partition NN module and an assembly NN module. For the new dataset, the partition NN module divides existing pre-trained NN models into blocks. The assembly NN module then selects some of these blocks and combines them into a new NN using a stitching component. To train PA-DNN, we design a score function that evaluates the performance of each new NN generated by PA-DNN. The evaluated value is used to train the partition NN module. Additionally, two loss functions are created to train the assembly NN module and the stitching component in the new NN, respectively. After the training process, PA-DNN infers a new NN, and only the stitching component of the NN is fine-tuned with the new dataset. Experiments show that, compared to manual models, neural architecture search, and the assembly model DeRy, PA-DNN can generate a more accurate and lightweight NN with lower training costs.

Multi-hop question answering over knowledge graph utilizes the knowledge graph (KG) structure to infer answers. However, KG often lacks edges in the reasoning path from the question entity to the answer entity. Recent research focused on various KG embedding methods to obtain the semantics of the reasoning path (called forward semantics) to repair missing edges. However, the forward semantics method could drift as the path get longer. This paper proposes a bidirectional semantics embedding and matching method (BSEM) to alleviate the forward semantics drift problem. BSEM first leverages a backward semantics method to deduce the semantics of the opposite direction of the reasoning path. Then, BSEM constructs a two-stage learning method to merge the bidirectional (forward or backward) semantics and find the correct answer. In the two-stage learning method, joint learning is created to learn the bidirectional semantics of the reasoning path simultaneously; contrast learning is also used to improve the ability of the backward semantics to identify the correct answers that are not found by the forward semantics. Experiments on the two benchmarks, MetaQA and WebQSP, show that BSEM surpasses the five baseline methods, PullNet, EmQL, LEGO, EmbedKGQA and KGT5. Especially for the incomplete KG – WebQSP, compared with the other four methods except for EmQL, BSEM improves the accuracy by 13.1%, 12.0%, 5.4% and 10.0%, respectively.

Genetic programming (GP) is a commonly used approach to solve symbolic regression (SR) problems. Compared with the machine learning or deep learning methods that depend on the pre-defined model and the training dataset for solving SR problems, GP is more focused on finding the solution in a search space. Although GP has good performance on large-scale benchmarks, it randomly transforms individuals to search results without taking advantage of the characteristics of the dataset. So, the search process of GP is usually slow, and the final results could be unstable. To guide GP by these characteristics, we propose a new method for SR, called Taylor genetic programming (TaylorGP)1. TaylorGP leverages a Taylor polynomial to approximate the symbolic equation that fits the dataset. It also utilizes the Taylor polynomial to extract the features of the symbolic equation: low order polynomial discrimination, variable separability, boundary, monotonic, and parity. GP is enhanced by these Taylor polynomial techniques. Experiments are conducted on three kinds of benchmarks: classical SR, machine learning, and physics. The experimental results show that TaylorGP not only has higher accuracy than the nine baseline methods, but also is faster in finding stable results.

Many recent studies focus on developing mechanisms to explain the black-box behaviors of neural networks (NNs). However, little work has been done to extract the potential hidden semantics (mathematical representation) of a neural network. A succinct and explicit mathematical representation of a NN model could improve the understanding and interpretation of its behaviors. To address this need, we propose a novel symbolic regression method for neural works (called SRNet) to discover the mathematical expressions of a NN. SRNet creates a Cartesian genetic programming (NNCGP) to represent the hidden semantics of a single layer in a NN. It then leverages a multi-chromosome NNCGP to represent hidden semantics of all layers of the NN. The method uses a (1+λ) evolutionary strategy (called MNNCGP-ES) to extract the final mathematical expressions of all layers in the NN. Experiments on 12 symbolic regression benchmarks and 5 classification benchmarks show that SRNet not only can reveal the complex relationships between each layer of a NN but also can extract the mathematical representation of the whole NN. Compared with LIME and MAPLE, SRNet has higher interpolation accuracy and trends to approximate the real model on the practical dataset1.