Identifying species in biology among tens of thousands of visually similar taxa while discovering unknown species in open-world environments remains a fundamental challenge in biodiversity research. Current methods treat identification and discovery as separate problems, with classification models assuming closed sets and discovery relying on threshold-based rejection. Here we present DeepTaxon, a retrieval-augmented multimodal framework that unifies species identification and discovery through interpretable reasoning over retrieved visual evidence. Given a query image, DeepTaxon retrieves the top-k candidate species with n exemplar images each from a retrieval index and performs chain-of-thought comparative reasoning. Critically, we redefine discovery as an explicit, retrieval-based decision problem rather than an implicit parametric memory problem. A sample is novel if and only if the retrieval index lacks sufficient evidence for identification, so each retrieval naturally yields a classification or discovery label without manual annotation, thereby providing automatic supervision for both tasks. We train the framework via supervised fine-tuning on synthetic retrieval-augmented data, followed by reinforcement learning on hard samples, converting high-recall retrieval into high-precision decisions that scale to massive taxonomic vocabularies. Extensive experiments on a large-scale in-distribution benchmark and six out-of-distribution datasets demonstrate consistent improvements in both identification and discovery. Ablation studies further reveal effective test-time scaling with candidate count k and exemplar count n, strong zero-shot transfer to unseen domains, and consistent performance across retrieval encoders, establishing an interpretable solution for biodiversity research.

Medical image classification plays a crucial role in disease diagnosis by assisting in disease analysis and treatment. However, many medical image datasets suffer from class imbalance due to the challenges of collecting sufficient image data in clinical and medical cases. Although numerous models demonstrate strong performance on balanced datasets, they often struggle when dealing with imbalanced classification tasks. To address this issue, we propose an imbalanced classification algorithm based on deep reinforcement learning, named DDQNbic. This algorithm is designed to prioritize the minority class by assigning higher rewards for correctly identifying underrepresented samples, thereby improving sensitivity to rare conditions. Building on this algorithm, we further develop BicDRL, an integrated framework specifically tailored for class-imbalanced medical image classification. We conduct extensive experiments on multiple image datasets, and the results demonstrate that our framework effectively leverages DRL to improve classification performance in imbalanced scenarios.

Advancements in neural networks have significantly enhanced the performance of classification models, achieving remarkable accuracy across diverse datasets. However, these models often lack transparency and do not support interactive reasoning with human users, which are essential attributes for applications that require trust and user engagement. To overcome these limitations, we introduce an innovative framework, Neural Causal Graph (NCG), that integrates causal inference with neural networks to enable interpretable and intervenable reasoning. We then propose an intervention training method to model the intervention probability of the prediction, serving as a contextual prompt to facilitate the fine-grained reasoning and human-AI interaction abilities of NCG. Our experiments show that the proposed framework significantly enhances the performance of traditional classification baselines. Furthermore, NCG achieves nearly 95\% top-1 accuracy on the ImageNet dataset by employing a test-time intervention method. This framework not only supports sophisticated post-hoc interpretation but also enables dynamic human-AI interactions, significantly improving the model's transparency and applicability in real-world scenarios.

Given the fact descriptions of legal cases, the legal judgment prediction (LJP) problem aims to determine three judgment tasks of law articles, charges, and the term of penalty. Most existing studies have considered task dependencies while neglecting the prior dependencies of labels among different tasks. Therefore, how to make better use of the information on the relation dependencies among tasks and labels becomes a crucial issue. To this end, we transform the text classification problem into a node classification framework based on graph reasoning and supervised contrastive learning (SCL) techniques, named GraSCL. Specifically, we first design a graph reasoning network to model the potential dependency structures and facilitate relational learning under various graph topologies. Then, we introduce the SCL method for the LJP task to further leverage the label relation on the graph. To accommodate the node classification settings, we extend the traditional SCL method to novel variants for SCL at the node level, which allows the GraSCL framework to be trained efficiently even with small batches. Furthermore, to recognize the importance of hard negative samples in contrastive learning, we introduce a simple yet effective technique called online hard negative mining (OHNM) to enhance our SCL approach. This technique complements our SCL method and enables us to control the number and complexity of negative samples, leading to further improvements in the model’s performance. Finally, extensive experiments are conducted on two well-known benchmarks, demonstrating the effectiveness and rationality of our proposed SCL approach as compared to the state-of-the-art competitors.

Drug combinations offer therapeutic benefits but also carry the risk of adverse drug-drug interactions (DDIs), especially under complex molecular structures. Accurate DDI event prediction requires capturing fine-grained inter-drug relationships, which are critical for modeling metabolic mechanisms such as enzyme-mediated competition. However, existing approaches typically rely on isolated drug representations and fail to explicitly model atom-level cross-molecular interactions, limiting their effectiveness across diverse molecular complexities and DDI type distributions. To address these limitations, we propose MolBridge, a novel atom-level joint graph refinement framework for robust DDI event prediction. MolBridge constructs a joint graph that integrates atomic structures of drug pairs, enabling direct modeling of inter-drug associations. A central challenge in such joint graph settings is the potential loss of information caused by over-smoothing when modeling long-range atomic dependencies. To overcome this, we introduce a structure consistency module that iteratively refines node features while preserving the global structural context. This joint design allows MolBridge to effectively learn both local and global interaction outperforms state-of-the-art baselines, achieving superior performance across long-tail and inductive scenarios. patterns, yielding robust representations across both frequent and rare DDI types. Extensive experiments on two benchmark datasets show that MolBridge consistently. These results demonstrate the advantages of fine-grained graph refinement in improving the accuracy, robustness, and mechanistic interpretability of DDI event prediction.This work contributes to Web Mining and Content Analysis by developing graph-based methods for mining and analyzing drug-drug interaction networks.

Hardware Trojans (HTs) are malicious modifications embedded in Integrated Circuits (ICs) that pose a significant threat to security. The concealment of HTs and the complexity of IC manufacturing make them difficult to detect. An effective solution is identifying HTs at the gate level through machine learning techniques. However, current methods primarily depend on end-to-end training, which fails to fully utilize the advantages of large-scale pre-trained models and transfer learning. Additionally, they do not take advantage of the extensive background knowledge available in massive datasets. This study proposes an HT detection approach based on large-scale pre-trained NLP models. We propose a novel approach named NtNDet, which includes a method called Netlist-to-Natural-Language (NtN) for converting gate-level netlists into a natural language format suitable for Natural Language Processing (NLP) models. We apply the self-attention mechanism of Transformer to model complex dependencies within the netlist. This is the first application of large-scale pre-trained models for gate-level netlists HT detection, promoting the use of pre-trained models in the security field. Experiments on the Trust-Hub, TRIT-TC, and TRIT-TS benchmarks demonstrate that our approach outperforms existing HT detection methods. The precision increased by at least 5.27%, The True Positive Rate (TPR) by 3.06%, the True Negative Rate (TNR) by 0.01%, and the F1 score increased by about 3.17%, setting a new state-of-the-art in HT detection.

The legal judgment prediction (LJP) task is to automatically decide appropriate law articles, charges, and term of penalty for giving the fact description of a law case. It considerably influences many real legal applications and has thus attracted theat tention of legal practitioners and AI researchers in recent years. In real scenarios, many confusing charges are encountered, which makes LJP challenging. Intuitively, for a controversial legal case, legal practitioners usually first obtain various possible judgment results as candidates based on the fact description of the case; then these candidates generally need to be carefully considered based on the facts and the rationality of the candidates. Inspiredby this observation, this paper presents a novel Recall&Rank-ing framework, dubbed asR2, which attempts to formalize LJPas a two-stage problem. The recall stage is designed to collect high-likelihood judgment results for a given case; these resultsare regarded as candidates for the ranking stage. The rankingstage introduces a verification technique to learn the relationships between the fact description and the candidates. It treats the par-tially correct candidates as semi-negative samples, and thus hasa certain ability to distinguish confusing candidates. Moreover,we devise a comprehensive judgment strategy to refine the finaljudgment results by comprehensively considering the rationality ofmultiple probable candidates. We carry out numerous experimentson two widely used benchmark datasets. The experimental resultsdemonstrate our proposed approach's effectiveness compared tothe other competitive baselines.

Detecting anomalies in logs is crucial for service and system management, since logs are widely used to record the run-time status, and are often the only data available for postmortem analysis. Since anomalies are usually rare in real-world services and systems, a common and feasible practice is to mine or learn normal patterns from logs, and deem those violating the normal patterns as anomalies. As log sequences are a kind of time series data, RNN (Recurrent Neural Network) and its variants have been extensively employed to capture the normal patterns. Nevertheless, the sequential nature of RNN and its variants makes them hard to parallelize and capture long-term dependencies, which may hinder their performance. To address this issue, in this paper we propose Loader, a novel semi-supervised log anomaly detector based on Transformer, because the Transformer architecture eschews recurrence and is able to draw global dependencies. Loader leverages the Transformer encoder to capture normal patterns from normal log sequences. When detecting, it gives a set of candidate log templates, that may appear after the input log substring under normal conditions. If the template of the actual next log message is not within the candidate set, this implies an anomaly. Previous similar methods select the most possible k log templates as candidates in any case, so the performance is sensitive to k, and it is nontrivial to pick a proper k. To alleviate this, we design a more flexible and robust 'top-p' algorithm, which determines the candidate set based on the cumulative probability of the most possible log templates. Extensive experiments are conducted based on three public log datasets, the experimental results validate the effectiveness and competitiveness of our approach.

Logs are pervasive in modern computing systems, and are valuable to service and system management. Nevertheless, with the rapidly growing size and complexity of computing systems, the log volume is exploding, which makes automatic log analysis imperative. Generally, in automatic log analysis, the first and fundamental step is log parsing, to which a lot of effort has been devoted. However, in most existing log parsing methods, log messages are merely treated as plain text. In natural language processing (NLP) area, it is a common practice to represent words and sentences with vectors, then the similarity between two words or sentences can be measured by the distance between their vectors. Inspired by these, we put forward a novel log parsing framework, named LPV ( L og P arser based on V ectorization), which performs log parsing by converting log messages and log templates into vectors, with the help of a vectorization method in NLP. LPV incorporates offline and online log parsing. In the offline log parsing, the central idea is to first represent log messages with vectors, so that the similarity between two log messages can be measured by the distance between their vectors, then we cluster log messages via clustering the vectors, and finally we extract log templates from the resultant clusters. By the end of the offline log parsing, each log template is assigned with an average vector, so that in the online log parsing, the similarity between an incoming log message and each log template can also be measured by the distance between their vectors. Extensive experiments have been conducted based on several public log datasets to evaluate LPV with three different vectorization methods. The results demonstrate that, with a proper vectorization method, LPV performs competitive with state-of-the-art log parsing methods, in both effectiveness and efficiency.

Effectively identifying compound-protein interactions (CPIs) is crucial for new drug design, which is an important step in silico drug discovery. Current machine learning methods for CPI prediction mainly use one-demensional (1D) compound/protein strings and/or the specific descriptors. However, they often ignore the fact that molecules are essentially modeled by the molecular graph. We observe that in real-world scenarios, the topological structure information of the molecular graph usually provides an overview of how the atoms are connected, and the local chemical context reveals the functionality of the protein sequence in CPI. These two types of information are complementary to each other and they are both significant for modeling compound-protein pairs. Motivated by this, we propose an end-to-end deep learning framework named GraphCPI, which captures the structural information of compounds and leverages the chemical context of protein sequences for solving the CPI prediction task. Our framework can integrate any popular graph neural networks for learning compounds, and it combines with a convolutional neural network for embedding sequences. To compare our method with classic and state-of-the-art deep learning methods, we conduct extensive experiments based on several widely-used CPI datasets. The experimental results show the feasibility and competitiveness of our proposed method.

The Gram-Schmidt orthogonalization algorithm and its related modified algorithms often show numerical instability when computing ill-conditioned or large-scale matrices.To solve this problem,this paper explores the cumulative effect of round-off errors of modified Gram-Schmidt algorithm(MGS),and then designs and implements a double-double precision modified Gram-Schmidt orthogonalization algorithm(DDMGS) based on the error-free transformation technology and double-double precision algorithm.A variety of accuracy tests illustrate that DDMGS algorithm has better numerical stability than the varients of BMGS_SVL,BMGS_CWY,BCGS_PIP and BCGS_PIO algorithms,which proves that DDMGS algorithm can effectively reduce the loss of orthogonality of matrix,improve the numerical accuracy,and demonstrate the stability of our algorithm.In the performance test, the floating point computations(flops) of different algorithms are calculated and then compared DDMGS algorithm with the modified Gram-Schmidt algorithm on ARM and Intel processors,the runtime of the DDMGS algorithm proposed in this paper isabout 5.03 and 18.06 times that of MGS respectively,but the accuracy is improved significantly.

With the rapid development of supercomputers, large-scale computing has become increasingly widespread in various scientific research and engineering fields. Meanwhile, the precision and efficiency of large-scale floating-point arithmetic have always been a research hotspot in high-performance computing. This paper studies the numerical method to solve large-scale sparse linear equations, in which the accumulation of rounding errors during the solution process leads to inaccurate results, and large-scale data makes the solver produce a long running time. For the above issues, we use error-free transformation technology and mixed-precision ideas to construct a reliable parallel numerical algorithm framework based on HYPRE, which solves large-scale sparse linear equations to improve accuracy and accelerate numerical calculations. Moreover, we illustrate the implementation details of our technique by implementing two cases. One is that we use error-free transformation technology to design high-precision iterative algorithms, such as GMRES, PCG, and BICGSTAB, which reduce rounding errors in the calculation process and make the result more accurate. The other is that we propose a mixed-precision iterative algorithm that utilizes low-precision formats to achieve higher computing power and reduce computing time. Experimental results demonstrate that XHYPRE has higher reliability and effectiveness . Our XHYPRE is on average 1.3x faster than HYPRE and reduces the number of iterations to 87.1% on average.

Singular value decomposition (SVD) plays a significant role in matrix analysis, and the differential quotient difference with shifts (DQDS) algorithm is an important technique for solving singular values of upper bidiagonal matrices. However, ill-conditioned matrices and large-scale matrices may cause inaccurate results or long computation times when solving singular values. At the same time, it is difficult for users to effectively find the desired solution according to their needs. In this paper, we design a precision-adjustable computational framework for solving singular values, named PACF. In our framework, the same solution algorithm contains three options: original mode, high-precision mode, and mixed-precision mode. The first algorithm is the original version of the algorithm. The second algorithm is a reliable numerical algorithm we designed using Error-free transformation (EFT) technology. The last algorithm is an efficient numerical algorithm we developed using the mixed-precision idea. Our PACF can add different solving algorithms for different types of matrices, which are universal and extensible. Users can choose different algorithms to solve singular values according to different needs. This paper implements the high-precision DQDS and mixed-precision DQDS algorithms and conducts extensive experiments on a supercomputing platform to demonstrate that our algorithm is reliable and efficient. Besides, we introduce the error analysis of the inner loop of the DQDS and HDQDS algorithms.

The cover tree (CT) and its improved version are hierarchical data structures that simplified navigating nets while maintaining good runtime guarantees. They can perform nearest neighbor search in logarithmic time and provide efficient computation in practice. In this article, we revisit cover trees for nearest neighbor search, and propose a more competitive method. The central idea of our method is to fully exploit the unused distance and direction information. More specially, our method introduces three novel concepts/techniques: (i) range list, (ii) quadrant information, and (iii) vectorial angle cosine. These techniques are seamlessly integrated into our suggested data structure and search algorithms. As an extra bonus, we explore approximate nearest neighbor and $k$ nearest neighbor based on the proposed techniques, and present algorithms for handling updates. Extensive experimental results, based on both real and synthetic datasets, consistently demonstrate that our method is attractive and competitive, compared against existing cover tree structures for nearest neighbor search and its variants.

本文设计并实现了面向深度学习的统一框架批处理矩阵乘法.我们细致地分析了利用矩阵乘法实现卷积的过程中卷积核、输入特征图和输出特征图在NCHW和NHWC两类存储格式下的矩阵数据排列特点,指出了其和矩阵行列主序的关系.在此基础上,为了更好复用共享的卷积核数据,我们提出将批量输入特征图转化为一个矩阵整体进行计算的方法.我们设计了统一框架的批处理分块矩阵乘法,该框架计算同一矩阵和多个不同矩阵的乘法,可以处理并输出任意存储格式的矩阵数据.我们优化了分块矩阵乘法实现,根据输入参数特征规划计算顺序,利用矩阵转置技巧复用核心计算模块,没有增加额外的数据组织操作.数值试验表明:本文设计实现的批处理单精度矩阵乘法的计算速度比循环调用原始单精度矩阵乘法的计算速度在处理中小尺度矩阵时在四款不同处理器平台上性能最高分别提高4.80％、26.57％、29.27％和25.55％,平均分别提升2.37％、14.37％、9.89％和15.72％.

Due to the cumulative effect of rounding errors, the Gram-Schmidt orthogonalization algorithm and its variants often exhibit numerical instability phenomena. Therefore, we analyze the data dependencies of the Modified Gram-Schmidt algorithm (MGS), and propose the compensated MGS algorithm (CompMGS) based on the idea of error-free transformation (EFT) technology and compensated algorithms. The theoretical analysis demonstrates that our algorithm has $\mathcal{O}(u)$ loss of orthogonality(in double working precision, $u=10^{-16}$ ) with the relaxed conditions $(\kappa_{2}(A) < 1^{-}0^{16})$ . A range of numerical experiments illustrate that CompMGS has similar effects to the double-double precision Modified Gram-Schmidt orthog-onalization algorithm (DDMGS) and is more accurate than the algorithms proposed in [1] and [2], which verifies the correctness of the theoretical analysis and proves the accuracy of our algorithm. While satisfying the same loss of orthogonality, the CompMGS is faster than DDMGS on both ARM and Intel processors by about 1.5 times and 2.3 times on average respectively, showing the high efficiency of our algorithm.

The Horner and Goertzel algorithms are frequently used in polynomial evaluation. Each of them can be less expensive than the other in special cases. In this paper, we present a new compensated algorithm to improve the accuracy of the Goertzel algorithm by using error-free transformations. We derive the forward round-off error bound for our algorithm, which implies that our algorithm yields a full precision accuracy for polynomials that are not too ill-conditioned. A dynamic error estimate in our algorithm is also presented by running round-off error analysis. Moreover, we show the cases in which our algorithms are less expensive than the compensated Horner algorithm for evaluating polynomials. Numerical experiments indicate that our algorithms run faster than the compensated Horner algorithm in those cases while producing the same accurate results, and our algorithm is absolutely stable when the condition number is smaller than 1016. An application is given to illustrate that our algorithm is more accurate than MATLAB’s fft function. The results show that the relative error of our algorithm is from 1015 to 1017, and that of the fft was from 1012 to 1015.

Aero-engine is the 'heart' of the aviation aircraft.Practical failure prediction of aero-engines is difficult due to the performance degradation covered by the continuous switching between various operating conditions.In order to solve the above problem, we propose a new type of aero-engine fault diagnosis model-RIRCNN (Residual Independently Reccurent and Convolutional Neural Network).It can process long sequences, and has superior feature extraction effect.We gather flight data sets through ground bench experiment of the aviation turboprop engine, and intensively conduct comparative experiments to evaluate the effectiveness of our model.The verification results demonstrate that our model can achieve excellent performance compared with other available baseline models.

The program cross-architecture migration technology can solve the problem that the software resources of the ARM architecture are not rich enough, so that the processors of the ARM architecture can obtain the abundant software resources of the x86 architecture. Since there are different migration schemes for different source programs, program migration technology involves many fields, and there is no unified method. This paper proposes a unified program migration framework model, and designs two technical modules of source code program migration and no-source code program migration according to whether the program has source code or not. The source code program migration module provides migration guidance for developers, reducing the difficulty of migration and improving the success rate of code migration; the no-source code migration module is based on binary translation technology, and integrates excellent open source tools QEMU and BOX64. Finally, experimental verification is carried out. The correctness and validity of the framework model.