Structured Electronic Health Record (EHR) data stores patient information in relational tables and plays a central role in clinical decision-making. Recent advances have explored the use of large language models (LLMs) to process such data, showing promise across various clinical tasks. However, the absence of standardized evaluation frameworks and clearly defined tasks makes it difficult to systematically assess and compare LLM performance on structured EHR data. To address these evaluation challenges, we introduce EHRStruct, a benchmark specifically designed to evaluate LLMs on structured EHR tasks. EHRStruct defines 11 representative tasks spanning diverse clinical needs and includes 2,200 task-specific evaluation samples derived from two widely used EHR datasets. We use EHRStruct to evaluate 20 advanced and representative LLMs, covering both general and medical models. We further analyze key factors influencing model performance, including input formats, few-shot generalisation, and finetuning strategies, and compare results with 11 state-of-the-art LLM-based enhancement methods for structured data reasoning. Our results indicate that many structured EHR tasks place high demands on the understanding and reasoning capabilities of LLMs. In response, we propose SEMaster, a code-augmented method that achieves state-of-the-art performance and offers practical insights to guide future research.

Alignment of large language models (LLMs) with human preferences typically relies on supervised reward models or external judges that demand abundant annotations. However, in fields that rely on professional knowledge, such as medicine and law, such large-scale preference labels are often unachievable. In this paper, we propose a generative entropy-guided preference modeling approach named GEM for LLMs aligment at low-resource and domain-specific scenarios. Instead of training a discriminative reward model on preference data, we directly train the LLM to internalize a closed-loop optimization architecture that can extract and exploit the multi-dimensional, fine-grained cognitive signals implicit in human preferences. Specifically, our \textit{Cognitive Filtering} module, based on entropy theory in decision making, first leverages Chain-of-Thought (CoT) prompting to generate diverse candidate reasoning chains (CoTs) from preference data. Subsequently, it introduces a token scoring mechanism to rank and weight the sampled CoTs, boosting the importance of high-confidence answers and strategically high-entropy tokens. Building on these filtered preferences, we fine-tune the LLM using a novel self-evaluated group advantage algorithm, \textit{SEGA}, which effectively aggregates group-level cognitive signals and transforms the entropy-based scores into implicit rewards for policy optimization. In these ways, GEM empowers the LLM to rely on its own judgments and establishes an entropy-guided closed-loop cognitive optimization framework, enabling highly efficient few-shot alignment of LLMs. Experiments on general benchmarks and domain-specific tasks (such as mathematical reasoning and medical dialogues) demonstrate that our GEM achieves significant improvements with few-shot preference data.

Anomaly detection aims to identify deviations from normal patterns within data. This task is particularly crucial in dynamic graphs, which are common in applications like social networks and cybersecurity, due to their evolving structures and complex relationships. Although recent deep learning-based methods have shown promising results in anomaly detection on dynamic graphs, they often lack of generalizability. In this study, we propose GeneralDyG, a method that samples temporal ego-graphs and sequentially extracts structural and temporal features to address the three key challenges in achieving generalizability: Data Diversity, Dynamic Feature Capture, and Computational Cost. Extensive experimental results demonstrate that our proposed GeneralDyG significantly outperforms state-of-the-art methods on four real-world datasets.

Misdiagnosis causes significant harm to healthcare systems worldwide, leading to increased costs and patient risks. MedRAG is a smart multimodal healthcare copilot equipped with powerful large language model (LLM) reasoning, designed to enhance medical decision-making. It supports multiple input modalities, including non-intrusive voice monitoring, general medical queries, and electronic health records. MedRAG provides recommendations on diagnosis, treatment, medication, and follow-up questioning. Leveraging retrieval-augmented generation enhanced by knowledge graph-elicited reasoning, MedRAG retrieves and integrates critical diagnostic insights, reducing the risk of misdiagnosis. It has been evaluated on both public and private datasets, outperforming existing models and offering more specific and accurate healthcare assistance. A demonstration video of MedRAG is available at: https://www.youtube.com/watch?v=PNIBDMYRfDM. The source code is available at: https://github.com/SNOWTEAM2023/MedRAG.

Retrieval-augmented generation (RAG) is a well-suited technique for retrieving privacy-sensitive Electronic Health Records (EHR).It can serve as a key module of the healthcare copilot, helping reduce misdiagnosis for healthcare practitioners and patients.However, the diagnostic accuracy and specificity of existing heuristic-based RAG models used in the medical domain are inadequate, particularly for diseases with similar manifestations.This paper proposes MedRAG, a RAG model enhanced by knowledge graph (KG)-elicited reasoning for the medical domain that retrieves diagnosis and treatment recommendations based on manifestations.MedRAG systematically constructs a comprehensive four-tier hierarchical diagnostic KG encompassing critical diagnostic differences of various diseases.These differences are dynamically integrated with similar EHRs retrieved from an EHR database, and reasoned within a large language model.This process enables more accurate and specific decision support, while also proactively providing follow-up questions to enhance personalized medical decision-making.MedRAG is evaluated on both a public dataset DDXPlus and a private chronic pain diagnostic dataset (CPDD) collected from Tan Tock Seng Hospital, and its performance is compared against various existing RAG methods.Experimental results show that, leveraging the information integration and relational abilities of the KG, our MedRAG provides more specific diagnostic insights and outperforms state-of-the-art models in reducing misdiagnosis rates.Our code will be available at https:// github.com/ SNOWTEAM2023/ MedRAG

Abstract Deep learning is a powerful technique to boost application performance in various fields, including face recognition, image classification, natural language understanding, and recommendation system. With the rapid increase in the computing power of mobile devices, developers can embed deep learning models into their apps for building more competitive products with more accurate and faster responses. Although there are several works of adversarial attacks against deep learning models in apps, they all need information about the models' internals (i.e., structures and weights) or need to modify the models. In this paper, we propose an effective black‐box approach by training substitute models to spoof the deep learning systems inside the apps. We evaluate our approach on 10 real‐world deep‐learning apps from Google Play to perform black‐box adversarial attacks. Through the study, we find three factors that can affect the performance of attacks. Our approach can reach a relatively high attack success rate of 66.60% on average. Compared with other adversarial attacks on mobile deep learning models, in terms of the average attack success rates, our approach outperforms its counterparts by 27.63%.

The small-loop time-domain electromagnetic (TDEM) method has been widely used in urban underground space detection in recent years due to the small workspace requirements. Due to the small side length of the transmitting coil of the small-loop electromagnetic method, more difficulties have been introduced in modeling and instrument development. The traditional modeling method cannot include source calculation, and the error becomes significantly large when calculating the initial field of the small-loop, and the cross iteration of electric and magnetic fields also increases the modeling error of the small-loop. Therefore, a high-precision 3-D small-loop source TDEM modeling method need to be proposed to provide the theoretical basis for feature analysis, inversion, and instrument parameter design. In this article, the electromagnetic wave equations are adopted as the controlling equations and discretized based on the Crank–Nicolson finite-difference time-domain (CN-FDTD) method. The entire computational space, including air and ground, is subdivided into sources to support 3-D small-loop TDEM modeling for shallow anomalous body conditions. Furthermore, the iterative formulas of the electromagnetic wave equations in the complex frequency-shifted perfect match layer (CFS-PML) are derived, the reflection errors are largely suppressed, and the modeling accuracy is significantly improved. Finally, the effectiveness of the improved method is verified by homogeneous models, layered models, and complex anomaly models. In addition, analyzing the propagation characteristics of the small-loop can guide the setting of the receiver sampling rate and improve the accuracy of detection. The results show that the improved method can achieve stable, low-memory, and high-precision 3-D small-loop source TDEM modeling, which can provide theoretical support for the application of the TDEM method in urban and shallow detection.

The induced polarization (IP) and superparamagnetic (SPM) multieffect fields are generated due to complex physical characteristics and parameter information of polymetallic particles in grounded-wire source time-domain electromagnetic (TDEM) detection. Recent studies show that the accurate observation of IP characteristics and elimination of SPM interference can improve TDEM interpretation accuracy effectively. However, it is difficult to realize the accurate recognition and effective observation of the multieffect responses in the large range of nonuniform grounded-wire source signals. Therefore, in this study, the 3-D modeling of IP and SPM effects in grounded-wire source TDEM is realized by introducing the fractional Cole-Cole conductivity and susceptibility models and establishing the double-curl electric field convolution matrices. The response characteristics of IP and SPM effects are analyzed, and a three-component observational method of multieffect is proposed. The effectiveness of the proposed method is verified by carrying out observation experiments of IP and SPM equivalent circuits. The results show that observing the magnetic field component parallel to the grounded-wire source can obtain IP and SPM response characteristics earlier, and observing the vertical component can reduce the interference of SPM effect while obtaining more obvious IP responses. Through the comprehensive analysis of three-component responses, the IP and SPM characteristics can be better determined. This study has a guiding role in receiving the information of multieffect responses and improving the accuracy of polymetallic ore detection.

With the improvement in the accuracy of time-domain electromagnetic method (TDEM) observations for geophysical exploration, abnormal diffusion phenomena have become a more salient focus of research. In particular, the induced polarization (IP) effect and magnetic viscosity (MV) effect are often observed in the exploration of polymetallic ore, with IP effect leading to a negative TDEM response, and MV effect generating a power-law decay of −0.6 to 1.4 in the late stage response. Ignoring these effects can lead to incorrect data interpretation. Therefore, to model and analyze both IP effect and MV effect accurately, a 3-D numerical modeling method for IP–MV effect is proposed. The Cole–Cole conductivity and Cole–Cole susceptibility models are approximated in the time domain using the multiple-zero-pole (MZP) method. Then, the diffusion equations for the electric and magnetization intensity fields are derived as control equations, while the 3-D modeling of the IP–MV effect with full waveform is realized based on the improved recursive convolution technique and finite-difference time-domain (FDTD) method. The effectiveness is verified by comparing with the 1-D numerical integration solutions. The response characteristics of IP effect and MV effect are analyzed with full trapezoidal waveform, and a complex model including IP–MV effect is discussed. It is demonstrated that IP–MV effect can be better observed at the ON-time stage. The proposed method can effectively model the diffusion process of IP–MV effect with the full waveform, which can be helpful for improving the inversion accuracy and detection precision for complex geological formations.

Recently, the electromagnetic (EM) anomalous diffusion phenomenon has been observed in time-domain EM (TDEM) surveys. Furthermore, the data interpretation accuracy has been reduced by adopting the traditional EM theory and methods. A number of models, such as random medium and roughness electrical conductivity theory, have been adopted to model the EM anomalous diffusion. However, problems such as modeling difficulty and massive discretization exist regarding characterizing the long-range correlation of EM anomalous diffusion. The space-fractional derivative has been proven to preferably describe the long-range correlation characteristic. Only a handful of studies on TDEM anomalous diffusion with space-fractional derivative have been conducted due to the difficulties in computational engineering problems. Therefore, we performed a series of studies about 2-D TDEM anomalous diffusion with space-fractional derivative. The 2-D TDEM space-fractional diffusion equation was constructed based on the space-fractional Ohm’s law model. Furthermore, the discretization and iteration forms of the control equation were derived based on the finite element method (FEM) by introducing the Riemann–Liouville (R–L)-type Riesz fractional derivatives. The 2-D mountain-shaped function and partial integration method (PIM) were combined to convert the fractional derivative into the primitive function form. Hence, the 2-D modeling of the space-fractional EM diffusion was realized. The effectiveness of our method was verified by the function construction method and wavenumber-domain analytical solution. The spatial and temporal characteristics of the space-fractional EM diffusion were analyzed by different geological models. Furthermore, we discuss the differences with the classical EM diffusion. Our method can effectively model the space-fractional EM diffusion in TDEM surveys and provide theoretical bases for improving the TDEM interpretation accuracy with complex geological conditions.

Search engines can quickly respond to a hyperlink list according to query keywords. However, when a query is complex, developers need to repeatedly refine search keywords and open a large number of web pages to find and summarize answers. Many research works of question and answering (Q&A) system attempt to assist search engines by providing simple, accurate, and understandable answers. However, without original semantic contexts, these answers lack explainability, making them difficult for users to trust and adopt. In this article, a brain-inspired search engine assistant named DeveloperBot based on knowledge graph is proposed, which aligns to the cognitive process of humans and has the capacity to answer complex queries with explainability. Specifically, DeveloperBot first constructs a multilayer query graph by splitting a complex multiconstraint query into several ordered constraints. Then, it models a constraint reasoning process as a subgraph search process inspired by a spreading activation model of cognitive science. In the end, novel features of the subgraph are extracted for decision-making. The corresponding reasoning subgraph and answer confidence are derived as explanations. The results of the decision-making demonstrate that DeveloperBot can estimate answers and answer confidences with high accuracy. We implement a prototype and conduct a user study to evaluate whether and how the direct answers and the explanations provided by DeveloperBot can assist developers' information needs.

The effect of magnetic viscosity (MV) has continued to be observed with the development and application of the time-domain electromagnetic method (TDEM). Field and laboratory data show that the MV effect is characterized by a -1±0.4 scope power-law delay in the late stage of electromagnetic response. Research on the MV effect can improve the detection accuracy of TDEM and assist in prospecting for ferromagnetic minerals. Most of the studies are based on the Chikazumi susceptibility model and the one-dimensional modeling method. However, the late-stage electromagnetic response shows a -1 power-law delay, which is inconsistent with the measured data. The log-uniform distribution of relaxation time τ in the Chikazumi model is not always appropriate. This study considers the Cole-Cole susceptibility model with a log-normal distribution of relaxation time. The three-dimensional (3D) modeling method of the MV effect is raised based on the rational function approximation algorithm and recursive convolution technique; the control equations and iterative process were adjusted based on finite-different time-domain (FDTD) method. The effectiveness was verified via half-space and layered models; the effects of susceptibility parameters on the response were clarified; moreover, the MV effect of the 3D anomalous model was analyzed. Our method can model the fractional propagation process of the MV effect more efficiently and help to improve the prospecting accuracy of the TDEM method under complex magnetic geological conditions.

API documentation provides important knowledge about the functionality and usage of APIs. In this paper, we focus on API caveats that developers should be aware of in order to avoid unintended use of an API. Our formative study of Stack Overflow questions suggests that API caveats are often scattered in multiple API documents, and are buried in lengthy textual descriptions. These characteristics make the API caveats less discoverable. When developers fail to notice API caveats, it is very likely to cause some unexpected programming errors. In this paper, we propose natural language processing(NLP) techniques to extract ten subcategories of API caveat sentences from API documentation and link these sentences to API entities in an API caveats knowledge graph. The API caveats knowledge graph can support information retrieval based or entity-centric search of API caveats. As a proof-of-concept, we construct an API caveats knowledge graph for Android APIs from the API documentation on the Android Developers website. We study the abundance of different subcategories of API caveats and use a sampling method to manually evaluate the quality of the API caveats knowledge graph. We also conduct a user study to validate whether and how the API caveats knowledge graph may improve the accessibility of API caveats in API documentation.

Knowledge graph is useful for many different domains like search result ranking, recommendation, exploratory search, etc. It integrates structural information of concepts across multiple information sources, and links these concepts together. The extraction of domain specific relation triples (subject, verb phrase, object) is one of the important techniques for domain specific knowledge graph construction. In this research, an automatic method named HDSKG is proposed to discover domain specific concepts and their relation triples from the content of webpages. We incorporate the dependency parser with rule-based method to chunk the relations triple candidates, then we extract advanced features of these candidate relation triples to estimate the domain relevance by a machine learning algorithm. For the evaluation of our method, we apply HDSKG to Stack Overflow (a Q&A website about computer programming). As a result, we construct a knowledge graph of software engineering domain with 35279 relation triples, 44800 concepts, and 9660 unique verb phrases. The experimental results show that both the precision and recall of HDSKG (0.78 and 0.7 respectively) is much higher than the openIE (0.11 and 0.6 respectively). The performance is particularly efficient in the case of complex sentences. Further more, with the self-training technique we used in the classifier, HDSKG can be applied to other domain easily with less training data.

Developers work in the IDE, but search online resources in the web browser. The separation of the working and search context often cause the ignorance of the working context during online search. Several tools have been proposed to integrate the web browser into the IDE so that developers can search and use online resources directly in the IDE. These tools enable only the shallow integration of the web browser and the IDE. Some tools allow the developer to augment search queries with program entities in the current snapshot of the code. In this paper, we present an in-IDE ambient search agent to bridge the separation of the developer's working context and search context. Our approach considers the developers' working context in the IDE as a time-series stream of programming event observed from the developer's interaction with the IDE over time. It supports the deeper integration of the working context in the entire search process from query formulation, custom search, to search results refinement and representation. We have implemented our ambient search agent and integrate it into the Eclipse IDE. We conducted a user study to evaluate our approach and the tool support. Our evaluation shows that our ambient search agent can better aid developers in searching and using online programming resources while working in the IDE.

With the explosive expansion of the information and communication technology (ICT) section, its energy saving has become an important issue and is receiving wide interest. In this study, we propose an adaptive clock frequency strategy for router cards to minimize the total energy consumption of an IP over WDM network. Rather than always running at full speed, the clock frequency of a router card is adaptively adjusted according to its actual-carried traffic demand. Given forecast traffic demand matrixes between different node pairs in different time slots, we develop a mixed integer linear programming (MILP) model to optimally choose the clock frequencies for each router card in different time slots such that the total energy consumption of the router cards is minimized. For lower computational complexity, the optimization model is also decomposed into two models, which correspond to the two subproblems of the optimization problem. The first subproblem minimizes the total number of router cards at each network node based on the peak-hour traffic, and the second subproblem optimally chooses the clock frequencies for each router card in different time slots. Due to the high-computational complexity of the MILP models, we also develop an efficient heuristic algorithm, in which two key steps that tackle the two subproblems are specifically developed. The joint MILP model provides a lower bound on the energy consumption, which shows to save more than 40% energy compared to the case without adaptive router card clock frequency. It is also found that the heuristic algorithm is efficient and performs close to the MILP models. In addition, the results also show that a router card supporting a small number of discrete clock frequencies can perform close to a card with continuously changed clock frequencies, and the benefit of adaptive clock frequency becomes weak with increasing router card power consumption overhead.