Learning decision policies from confounded observational data is a challenging task in causal inference, as unobserved confounders can lead to biased or suboptimal actions when relying solely on machine learning models. A synergistic approach is learning to defer, which decides when to act itself and when to defer to a human expert with access to unobserved information. However, constructing the learning target, which defines the probability of choosing each action or deferral, remains a core challenge. To address this, we propose causal-target-based learning to defer (CTLD) framework, where the causal target is constructed from sharp bounds on potential outcomes. Specifically, the degree of overlap between these bounds determines the probability of deferral, while their relative positions and widths define the probabilities over actions. CTLD aligns model predictions with this causal target to make probabilistic decisions over actions and deferral. We present comprehensive theoretical guarantees for the learned policy and demonstrate the effectiveness of CTLD on synthetic and semi-synthetic datasets.

Using Web Mining and Content Analysis to find and understand clues from the massive amount of unstructured text online is very important for predicting future events and providing early risk warnings in important fields like finance and public safety. While Large Language Models (LLMs) exhibit potential in processing and understanding text, current text-based event prediction faces two primary challenges: first, an insufficient utilization of potential information within the text, such as causal relationships and latent associations, and second, limited predictive reliability constrained by issues like the LLM's own ability and hallucinations. To address these challenges, we propose a novel event prediction framework, Bidirectional Reasoning with Self-Correction (BRSC). BRSC comprises two complementary reasoning dimensions: temporal deductive reasoning, which analyzes the trajectory of historical events along the timeline to enable accurate trend extrapolation, and synchronic associative reasoning, which deeply mines details and latent connections from documents within a specific time window to extended semantic information. In addition, we use a self-correction mechanism that identifies and rectifies potential hallucinations and errors during the reasoning process. Extensive experiments on international relations event prediction demonstrate that BRSC achieves significant improvements over several leading LLM-based methods.

• Random Forest achieved AUC 0.97 for melanoma detection using 1064 nm features. • Weakly supervised denoising cut false positives by 42%, raising accuracy to 94%. • State-space PAFCMamba outperformed Transformer, with AUC 0.785 versus 0.730. • Hybrid CNN-Transformer predicted staging & metastasis with AUC up to 0.974. • PAFC spectra linked to immune genes, enabling noninvasive patient stratification. Melanoma remains one of the most aggressive forms of skin cancer, with early detection being critical for patient outcomes. This study introduces a novel photoacoustic fingerprinting approach integrated with advanced machine learning for non-invasive melanoma detection and circulating tumor cell (CTC) identification. We developed a three-tiered photoacoustic fingerprinting system combining photoacoustic flow cytometry (PAFC) with machine learning algorithms. A uniform PAFC configuration employed a 532 nm laser for vascular localization followed by a 1064 nm laser targeting melanin-rich melanoma cells. The study included 50 melanoma patients and healthy controls, analyzing spectral features across multiple wavelengths. We compared self-supervised learning architectures (PAFCMamba vs. Transformer) and developed a hybrid CNN-Transformer model for simultaneous CTC identification, staging, and metastatic dissemination prediction. The photoacoustic fingerprinting system achieved exceptional diagnostic discrimination between melanoma patients and healthy controls. Random Forest achieved area under curve (AUC) values up to 0.97. The PAFCMamba model outperformed the Transformer architecture (accuracy 0.75 vs. 0.62, AUC 0.785 vs. 0.730). The hybrid CNN-Transformer architecture achieved exceptional performance with AUCs up to 0.974 and precision > 94 % in simultaneous CTC detection, staging, and metastasis prediction. High-immunogenicity genes including MLANA, GPR89B/A, and PIGF were identified as potential immunotherapy targets, with photoacoustic signatures serving as non-invasive surrogate biomarkers for underlying molecular characteristics. This study establishes photoacoustic fingerprinting as a clinically viable, non-invasive approach for melanoma detection and CTC monitoring, achieving performance comparable to conventional methods. The integration of machine learning with photoacoustic biomarkers provides a scalable framework with interpretable features that facilitates clinical translation.

Onboard IoT edge computing provides a real-time and efficient platform for levitation control of the high-speed maglev train. Aiming at the stability control problem of levitation systems, a closed-loop control framework integrating Transformer networks with Model Predictive Control (MPC) to enhance levitation stability and dynamic response performance is proposed. To handle time-varying dynamics and disturbances, an edge-deployed temporal Transformer network predicts short-horizon system states, improving MPC optimization within finite prediction horizons. Simultaneously, safety constraints are introduced with the combination of Control Barrier Function (CBF) method, ensuring critical variables, including levitation gap, remain within safe bounds. The framework covers nonlinear dynamics, linearized system equations, Transformer-based forecasting, MPC design with constraints, and CBF-embedded safety control. Experiments on the hardware-in-the-loop test platform demonstrate significant performance improvements over traditional methods: maximum deviation reduced by 50%, recovery time shortened by 60%, and response speed increased by 50%, while maintaining safety during high-speed operation. However, these results were obtained in a simulated environment rather than under actual operating conditions at 600 km/h, and the effects of aerodynamic disturbances at 600 km/h still require further investigation in subsequent research.

Large language models (LLMs) demonstrate remarkable capabilities in understanding complex tasks and have achieved commendable performance in graph-related tasks, such as node classification, link prediction, and subgraph classification.These tasks primarily depend on the local reasoning capabilities of the graph structure.However, research has yet to address the graph partitioning task that requires global perception abilities.Our preliminary findings reveal that vanilla LLMs can only handle graph partitioning on extremely small-scale graphs.To overcome this limitation, we propose a three-phase pipeline to empower LLMs for large-scale graph partitioning: coarsening, reasoning, and refining.The coarsening phase reduces graph complexity.The reasoning phase captures both global and local patterns to generate a coarse partition.The refining phase ensures topological consistency by projecting the coarse-grained partitioning results back to the original graph structure.Extensive experiments demonstrate that our framework enables LLMs to perform graph partitioning across varying graph scales, validating both the effectiveness of LLMs for partitioning tasks and the practical utility of our proposed methodology.