Large Language Models (LLMs) suffer from order bias, where their performance is affected by the arrangement order of input elements.This unfairness limits the model's applications in scenarios such as in-context learning and Retrieval-Augmented Generation (RAG).Recent studies attempt to obtain optimal or suboptimal arrangements based on statistical results or using dataset-based search, but these methods increase inference overhead while leaving the model's inherent order bias unresolved.Other studies mitigate order sensitivity through supervised fine-tuning using augmented training sets with multiple order variants, but often at the cost of accuracy, trapping the model in consistent yet incorrect hallucinations.In this paper, we propose Dual Group Advantage Optimization (DGAO), which aims to improve model accuracy and order stability simultaneously.DGAO calculates and balances intragroup relative accuracy advantage and intergroup relative stability advantage, rewarding the policy model for generating order-stable and correct outputs while penalizing ordersensitive or incorrect responses.This marks the first time reinforcement learning has been used to mitigate LLMs' order sensitivity.We also propose two new metrics, Consistency Rate and Overconfidence Rate, to reveal the pseudostability of previous methods and guide more comprehensive evaluation.Extensive experiments demonstrate that DGAO achieves superior order fairness while improving performance on RAG, mathematical reasoning, and classification tasks.Our

High-stakes domains such as finance, law, and biomedicine demand both accurate results and rigorous reasoning.Current reinforcement learning paradigms primarily rely on outcomebased rewards, often overlooking latent logical fallacies in intermediate steps.Leveraging the cognitive asymmetry where falsifying local errors is more efficient than generating global correctness, we propose RADO (Reasoning Audit-Driven Optimization).RADO introduces a specialized audit model augmented with external tools to identify local logical ruptures and calibrate reward signals.By integrating Direct Preference Optimization (DPO) with Group Relative Policy Optimization (GRPO), our framework enables explicit supervision over reasoning paths.Experimental results demonstrate that RADO consistently improves final accuracy while significantly enhancing logical rigor in high-stakes domains.

Graph-related tasks are traditionally addressed with Graph Neural Networks (GNNs) or graph transformers, but their task-specific training limits generalization.Large Language Models (LLMs) offer stronger generalization, yet encoding graphs as one-dimensional text struggles to capture multi-hop dependencies and two-dimensional topology.Vision-Language Models (VLMs) provide an alternative by visualizing graphs, but rendering large graphs in a single image causes clutter, occlusion, and distraction, hindering reasoning.We propose MuSe, a novel multi-stage graph reasoning framework based on VLMs.Instead of processing entire graphs at once, MuSe incrementally samples and visualizes task-relevant subgraphs, enabling progressive reasoning.The framework employs a two-stage training paradigm: supervised fine-tuning to acquire local sampling and reasoning skills, followed by reinforcement learning with GRPO to refine the sampling strategy and control dialog length.To support evaluation, we introduce LGVLQA, a new multimodal dataset with larger and more complex graph structures, addressing the scalability limitations of existing benchmarks.Experiments show that MuSe consistently outperforms leading LLM and VLM baselines, demonstrating improved structural understanding and reasoning ability.Our code and data are available at this url.

In the era of transformative artificial intelligence (AI), professional Master's programs must navigate both technological breakthroughs and pedagogical shifts. This paper presents an innovative approach adopted by the College of Software and Microelectronics at Peking University to train Electronic Information Master's students through a multi-dimensional framework encompassing academic research, industry applications, and in-service capacity building. By transitioning from code-focused instruction to a higher-level AI toolchain paradigm, the curriculum leverages large-model technologies (e.g., GPT, Chat-GPT) to promote greater efficiency in problem-solving, project-based learning, and critical thinking. The New Engineering Experimental Class exemplifies how enterprises partner with academia to guide real-world project cycles under dual mentor-ship, thereby accelerating technology transfer and strengthening student competencies in both theory and practice. In parallel, a three- tiered engineering practice system-ranging from course-embedded labs to full-scale industrial internships-amplifies hands-on development and integrates rigorous assessments of engineering proficiency. Furthermore, by incorporating cross-disciplinary mentorships, AI-enhanced campus tools, and cutting-edge lectures on new engineering and data-driven domains, the program produces graduates who are well-positioned in algorithm development, system architecture, and technology management. These reforms demonstrate an effective means of aligning educational supply with market demand, empowering students to address the evolving challenges of AI -driven industries while championing ethical and sustainable practices. The paper concludes by discussing the program's adaptability for continued innovation amid AGI-frontier breakthroughs and digital transformation imperatives.

Dynamic graph augmentation is used to improve the performance of dynamic GNNs. Most methods assume temporal locality, meaning that recent edges are more influential than earlier edges. However, for temporal changes in edges caused by random noise, overemphasizing recent edges while neglecting earlier ones may lead to the model capturing noise. To address this issue, we propose STAA (SpatioTemporal Activity-Aware Random Walk Diffusion). STAA identifies nodes likely to have noisy edges in spatiotemporal dimensions. Spatially, it analyzes critical topological positions through graph wavelet coefficients. Temporally, it analyzes edge evolution through graph wavelet coefficient change rates. Then, random walks are used to reduce the weights of noisy edges, deriving a diffusion matrix containing spatiotemporal information as an augmented adjacency matrix for dynamic GNN learning. Experiments on multiple datasets show that STAA outperforms other dynamic graph augmentation methods in node classification and link prediction tasks.

Current popular Large Vision-Language Models (LVLMs) are suffering from Hallucinations on Object Attributes (HoOA), leading to incorrect determination of fine-grained attributes in the input images. Leveraging significant advancements in 3D generation from a single image, this paper proposes a novel method to mitigate HoOA in LVLMs. This method utilizes multiview images sampled from generated 3D representations as visual prompts for LVLMs, thereby providing more visual information from other viewpoints. Furthermore, we observe the input order of multiple multiview images significantly affects the performance of LVLMs. Consequently, we have devised Multiview Image Augmented VLM (MIAVLM), incorporating a Multiview Attributes Perceiver (MAP) submodule capable of simultaneously eliminating the influence of input image order and aligning visual information from multiview images with Large Language Models (LLMs). Besides, we designed and employed negative instructions to mitigate LVLMs’ bias towards "Yes" responses. Comprehensive experiments demonstrate the effectiveness of our method.

Multimedia Event Extraction (MEE) has become an important task in information extraction research as news today increasingly prefers to contain multimedia content. Current MEE works mainly face two challenges: (1) Inadequate extraction framework modeling for handling complex and flexible multimedia event structure; (2) The absence of multimodal-aligned training data for effective knowledge transfer to MEE task. In this work, we propose a Stepwise Schema-Guided Prompting Framework (SSGPF) using Multimodal Large Language Model (MLLM) as backbone for adaptive structure capturing to solve MEE task. At the initial step of SSGPF, we design Event Type Schema Guided Prompting (ETSGP) for event detection, then we devise Argument Role Schema Guided Prompting (ARSGP) that contains multi-step prompts with text-bridged grounding technique for argument extraction. We construct a weakly-aligned multimodal event labeled dataset based on existing unimodal event annotations, then conduct parameter efficient instruction tuning with LoRA on LLaVA-v1.5-7B under SSGPF. Experiments on the M2E2 benchmark demonstrate that SSGPF significantly outperforms current SOTA baselines by 5.8 percent F1 on event detection and 8.4 percent F1 on argument extraction.

This study addresses the challenge of ambiguity in knowledge graph question answering (KGQA). While recent KGQA systems have made significant progress, particularly with the integration of large language models (LLMs), they typically assume user queries are unambiguous, which is an assumption that rarely holds in real-world applications. To address these limitations, we propose a novel framework that dynamically handles both entity ambiguity (e.g., distinguishing between entities with similar names) and intent ambiguity (e.g., clarifying different interpretations of user queries) through interactive clarification. Our approach employs a Bayesian inference mechanism to quantify query ambiguity and guide LLMs in determining when and how to request clarification from users within a multi-turn dialogue framework. We further develop a two-agent interaction framework where an LLM-based user simulator enables iterative refinement of logical forms through simulated user feedback. Experimental results on the WebQSP and CWQ dataset demonstrate that our method significantly improves performance by effectively resolving semantic ambiguities. Additionally, we contribute a refined dataset of disambiguated queries, derived from interaction histories, to facilitate future research in this direction.1

Compositional Zero-Shot Learning (CZSL) aims at recognizing novel compositions by combining concepts learned from seen compositions. The key to tackle CZSL is disentangling highly coupled attribute-object compositions and learning exclusive concepts. Previous works mainly design networks to learn visual concepts from top-layer representations provided by visual backbones. As visual backbones progressively integrate information layer by layer, some low-level but critical information for concept learning may be lost, and the coupling between attribute and object features deepens. To address these issues, we propose to extract multi-scale visual features and fuse them in an adaptive way by Mixture of Experts (MoE) networks. We also employ feature-level similarity and a maximum entropy regularization term to constrain the model to effectively disentangle and learn concepts from multi-scale visual information. Comprehensive experiments on three CZSL benchmark datasets demonstrate that our method significantly outperforms previous SOTA methods in both closed-world and open-world settings.

In recent years, with the rapid development of artificial intelligence technology, new services represented by the combination of large models and fields have become a strong driving force for social, technological, and economic development. The new situation has brought tremendous changes to the technical connotation of service computing, and has also brought new requirements and challenges to the cultivation of professional talents in service computing. Based on the objective laws of talent cultivation, the School of Software and Microelectronics at Peking University has conducted a series of explorations and innovations in the teaching practice of service computing, targeting the characteristics of the engineering field and the unique needs of new engineering disciplines. By setting up specialized research directions and using a modular curriculum system as the content support for targeted teaching, and through a series of characteristic training links integrating industry and education, a trinity training model of “direction content process” has been formed. This training model is highly popular among students, and the popularity of enrollment and application has been increasing year by year. The talents cultivated have also received praise from the industry.

The recommendation of suitable hashtags for mi-croposts encompassing multimodal content stands as a pivotal challenge for numerous Social Networking Service (SNS) applications such as Instagram, Weibo, etc. The accuracy of multimodal hashtag recommendation algorithms relies heavily on the comprehension of multimodal information, user historical information, and the reasoning ability based on such information. However, most previous works have not effectively utilized both historical and additional information simultaneously. Large Language Models (LLMs) learn a vast amount of implicit knowledge during the pre-training stage, which can serve as potential knowledge bases while also possessing strong reasoning abilities. Therefore, LLMs can provide additional information to help understand the micropost content and infer suitable hashtags with strong reasoning ability. However, introducing LLMs for multimodal hashtag recommendation faces three main challenges. Firstly, LLMs require an efficient modality alignment module to accept a multimodal input. Secondly, LLMs are highly sensitive to input order, while utilizing user historical information requires accepting multiple historical samples, necessitating the design of a robust historical information processing module to eliminate the influence of input order. Thirdly, fine-tuning LLMs entails substantial computational overheads, necessitating the reduction of additional trainable parameters. To address the first two challenges, this paper designs an efficient modality alignment module capable of processing multiple historical samples, simultaneously addressing the sensitivity of LLMs to input order changes. To tackle the third challenge, a hybrid prompt learning approach utilizing both soft and hard prompts is proposed to achieve parameter-efficient fine-tuning of LLMs. Finally, a LLM-augmented Multimodal Hashtag Recommendation algorithm (LLM-MHR) is implemented. Comprehensive experiments on the representative dataset MACON demonstrate that LLM-MHR has achieved SOTA performances with significant improvements.

Image-text retrieval, a fundamental cross-modal task, performs similarity reasoning for images and texts. The primary challenge for image-text retrieval is cross-modal semantic heterogeneity, where the semantic features of visual and textual modalities are rich but distinct. Scene graph is an effective representation for images and texts as it explicitly models objects and their relations. Existing scene graph based methods have not fully taken the features regarding various granularities implicit in scene graph into consideration (e.g., triplets), the inadequate feature matching incurs the absence of non-trivial semantic information (e.g., inner relations among triplets). Therefore, we propose a S emantic-Consistency E nhanced M ulti-Level Scene Graph Matching (SEMScene) network, which exploits the semantic relevance between visual and textual scene graphs from fine-grained to coarse-grained. Firstly, under the scene graph representation, we perform feature matching including low-level node matching, mid-level semantic triplet matching, and high-level holistic scene graph matching. Secondly, to enhance the semantic-consistency for object-fused triplets carrying key correlation information, we propose a dual-step constraint mechanism in mid-level matching. Thirdly, to guide the model to learn the semantic-consistency of matched image-text pairs, we devise effective loss functions for each stage of the dual-step constraint. Comprehensive experiments on Flickr30K and MS-COCO datasets demonstrate that SEMScene achieves state-of-the-art performances with significant improvements.

Event detection aims to detect events from the text by identifying and classifying event triggers (the most representative words). Most of the existing works rely heavily on complex downstream networks and require sufficient training data. Thus, those models may be structurally redundant and perform poorly when data is scarce. Prompt-based models are easy to build and are promising for few-shot tasks. However, current prompt-based methods may suffer from low precision because they have not introduced event-related semantic knowledge (e.g., part of speech, semantic correlation, etc.). To address these problems, this paper proposes a Knowledge-injected Prompt Tuning (KiPT) model. Specifically, the event detection task is formulated into a condition generation task. Then, knowledge-injected prompts are constructed using external knowledge bases, and a prompt tuning strategy is leveraged to optimize the prompts. Extensive experiments indicate that KiPT outperforms strong baselines, especially in few-shot scenarios.

Answering natural language questions on knowledge graphs (KGQA) remains a great challenge in terms of understanding complex questions via multi-hop reasoning. Previous efforts usually exploit large-scale entity-related text corpus or knowledge graph (KG) embeddings as auxiliary information to facilitate answer selection. However, the rich semantics implied in off-the-shelf relation paths between entities is far from well explored. This paper proposes improving multi-hop KGQA by exploiting relation paths’ hybrid semantics. Specifically, we integrate explicit textual information and implicit KG structural features of relation paths based on a novel rotate-and-scale entity link prediction framework. Extensive experiments on three existing KGQA datasets demonstrate the superiority of our method, especially in multi-hop scenarios. Further investigation confirms our method’s systematical coordination between questions and relation paths to identify answer entities.

Many recent sentence-level event detection efforts focus on enriching sentence semantics, e.g., via multi-task or prompt-based learning. Despite the promising performance, these methods commonly depend on label-extensive manual annotations or require domain expertise to design sophisticated templates and rules. This paper proposes a new paradigm, named dialogue-based explanation, to enhance sentence semantics for event detection. By saying dialogue-based explanation of an event, we mean explaining it through a consistent information-intensive dialogue, with the original event description as the start utterance. We propose three simple dialogue generation methods, whose outputs are then fed into a hybrid attention mechanism to characterize the complementary event semantics. Extensive experimental results on two event detection datasets verify the effectiveness of our method and suggest promising research opportunities in the dialogue-based explanation paradigm.

With the help of the sufficiency of heterogeneous information, heterogeneous information network(HIN) has been treated as the most advanced method to extract complex semantic data in recommender system. But it is still an empty field for some traditional industries such as testing and inspection industry, which mainly adopt the similarity-based collaborative filtering(CF) method. But it will make a huge waste of the rich heterogeneous auxiliary data, which could be fully utilized by HIN based method. Especially for testing and inspection industry, the profession will help the model to find a more accurate match between the user and business. In this work, we succeeded in building up a HIN embedding approach for recommendation, and design a unique network structure for testing and inspection industry, which both utilize the rich underlying information and properly solve the specialty problem in a professional industry, different from normal recommender scenario. An intensive experiment on the real world data set shows the performance of the model.

Beam search is a popular algorithm for solving real-world problems --- especially where search space is an enormously large tree but real-time solutions are most preferred. We present a memory-bounded best-first beam search (MB2FBS), which can be viewed as an improved and generalized version of standard beam search in trees. The algorithm takes three parameters --- in contrast to the singular parameter beam size in standard beam search. We discuss how to recover standard beam search and how to realize other search behavior by setting these three parameters correspondingly. In particular, we show that the principal version of MB2FBS can be thought as an algorithm whose search expense is similar or upper bounded by beam search of certain beam size; however it often finds better solutions as it decides the number of nodes to be searched each depth dynamically with respect cost landscape. We apply our algorithm for tensor program auto-scheduling in Halide, an important industrial problem that uses tree search for optimizing tensor program executions. We show that the principal variants of MB2FBS deliver better empirical results than the highly optimized beam search counterpart. Most importantly, it finds superior schedules while no more computation cost is used for search, which is highly desirable for real-time program compilation and optimization.

Efficient execution is crucial to the successful deployment of deep learning models to real-world applications. Considerable recent effort have been devoted to computer systems for automatic discovery of efficient schedules for executing tensor programs on given hardware platforms. Built on TVM [1], Ansor [2] is the most recent and state-of-the-art framework for auto-scheduling deep neural net computation pipelines. In this paper, we present Bansor, an improved version of Ansor for automatic optimization of tensor programs, using bandit-based reinforcement learning (RL). We reexamine the algorithmic procedures of Ansor, and identify two selection sub-problems where RL techniques are readily usable. We then introduce bandit-based algorithms to balance exploration and exploitation during the processes of sketch selection and task scheduling. We evaluate the resulting algorithm, Bansor, on a wide range of tensor programs and two hardware platforms. Experiment results show that Bansor yields significant improvement over Ansor. On hard network test cases, Bansor uses an order of magnitude less number of measurement trails to attain Ansor’s best schedules, eventually converging to significantly better results given an equal number of measurement trials, despite the fact that Ansor’s performance has been superb.

As the core of quantitative trading, indicator effectiveness continuously plays a vital role in stock prediction. The majority of studies are currently dedicated to constructing indicators with high Pearson Correlation Coefficient (CORR) with returns. However, the pursuit of high CORR may ignore some indicators that produce high profits. Therefore, we propose a new Bilateral Correlation Coefficient (BCORR), which can detect some profitable indicators that are previously discarded due to low CORR. BCORR is a weighted correlation coefficient, and the weight is a variable related to the return so that the top and bottom ranked returns have a more significant impact on it. To generate an indicator that has high BCORR with the return, we propose a framework called the Bilateral Autotrading Framework (BAF) based on Bilateral Loss to forecast the cross-sectional rank of stock return, and the prediction is adopted as a bilateral indicator to select stocks to invest. Meanwhile, the positions of the selected stocks are optimized by the Sharpe-oriented optimization to reduce the risk and improve the return. Experiments on real-world stock market datasets show that the BAF can significantly improve performance to deep stock prediction methods, such as Transformer, LSTM, and NBEATS.

Natural Language Sentence Matching (NLSM) serves as the core of many natural language processing tasks.1) Most previous work develops a single specific neural model for NLSM tasks.2) There is no previous work considering adversarial attack to improve the performance of NLSM tasks.3) Adversarial attack is usually used to generate adversarial samples that can fool neural models.In this paper, we first find a phenomenon that different categories of samples have different vulnerabilities.Vulnerability is the difficulty degree in changing the label of a sample.Considering the phenomenon, we propose a general two-stage training framework to enhance neural models with Vulnerability via Adversarial Attack (VAA).We design criteria to measure the vulnerability which is obtained by adversarial attack.VAA framework can be adapted to various neural models by incorporating the vulnerability.In addition, we prove a theorem and four corollaries to explain the factors influencing vulnerability effectiveness.Experimental results show that VAA significantly improves the performance of neural models on NLSM datasets.The results are also consistent with the theorem and corollaries.The code is released on https://github.com/rzhangpku/VAA.

Similar case matching (SCM) aims to determine whether legal case documents are similar or not. In fact, SCM is an extension of the semantic text matching. Various deep learning models are proposed to solve the semantic text matching problems. However, the main difference between the case documents may be subtle, and the length of documents can be quite long. Moreover, the case documents are written in structural format and contain plenty of legal terms. To address these challenges, we propose a novel model in this paper. Accordingly, the legal feature vector is introduced into the semantic text matching model, and BERT is adopted as the encoding layer to capture long-range dependencies in the case documents. We conduct several experiments to evaluate the performance of our proposed model. The results show that our model outperforms other existing methods on the public dataset CAIL2019-SCM.

Deep neural networks have recently become a popular solution to keyword spotting systems, which enable the control of smart devices via voice. In this paper, we apply neural architecture search to search for convolutional neural network models that can help boost the performance of keyword spotting based on features extracted from acoustic signals while maintaining an acceptable memory footprint. Specifically, we use differentiable architecture search techniques to search for operators and their connections in a predefined cell search space. The found cells are then scaled up in both depth and width to achieve competitive performance. We evaluated the proposed method on Google's Speech Commands Dataset and achieved a state-of-the-art accuracy of over 97% on the setting of 12-class utterance classification commonly reported in the literature.

Duplicate Question Identification (DQI) improves the processing efficiency and accuracy of large-scale community question answering and automatic QA system. The purpose of DQI task is to identify whether the paired questions are semantically equivalent. However, how to distinguish the synonyms or homonyms in paired questions is still challenging. Most previous works focus on the word-level or phrase-level semantic differences. We firstly propose to explore the asking emphasis of a question as a key factor in DQI. Asking emphasis bridges semantic equivalence between two questions. In this paper, we propose an attention model with multi-fusion asking emphasis (MFAE) for DQI. At first, BERT is used to obtain the dynamic pre-trained word embeddings. Then we get inter- and intra-asking emphasis by summing inter-attention and self-attention, respectively; the idea is that, the more a word interacts with others, the more important the word is. Finally, we use eight-way combinations to generate multi-fusion asking emphasis and multi-fusion word representation. Experimental results demonstrate that our model achieves state-of-the-art performance on both Quora Question Pairs and CQADupStack data. In addition, our model can also improve the results for natural language inference task on SNLI and MultiNLI datasets. The code is available at https://github.com/rzhangpku/MFAE.

To improve the efficiency of elderly assessments, an influence-based fast preceding questionnaire model (FPQM) is proposed. Compared with traditional assessments, the FPQM optimizes questionnaires by reordering their attributes. The values of low-ranking attributes can be predicted by the values of the high-ranking attributes. Therefore, the number of attributes can be reduced without redesigning the questionnaires. A new function for calculating the influence of the attributes is proposed based on probability theory. Reordering and reducing algorithms are given based on the attributes’ influences. The model is verified through a practical application. The practice in an elderly-care company shows that the FPQM can reduce the number of attributes by 90.56% with a prediction accuracy of 98.39%. Compared with other methods, such as the Expert Knowledge, Rough Set and C4.5 methods, the FPQM achieves the best performance. In addition, the FPQM can also be applied to other questionnaires.

In order to improve the effect and the efficiency of the mechanical ventilation in grain storage, a context aware method is proposed. The context meta-model is defined which builds the basis for context modeling. The context model for mechanical ventilation in grain storage is built, which had full consideration of the ventilation purpose and the lifecycle of grain storage. The feasibility and the correctness of the model is verified with the context aware service system, which can receive the context information of the granary to figure out the real situation and make the ventilation plan accordingly.

We have deployed a big data platform for the insurance company. One can select the data they need from database, do data analysis job and save the final model to the model pool using the platform. We completed churn prediction task on both SPSS [1] and Spark [7] using the data providing by X insurance company and carefully compared the execution flow, runtime, model evaluation, and model precision of each. Experimental results confirm that Spark ML [6] is easy to use and can cope with big data problems.

The research on service engineering will promote the productivity, effectiveness, and efficiency of service system and service industry as well. This paper gives an overview of Service Engineering. The paper starts form Service logy, which is believed a new emerging discipline. We put forward the concept of service logy and describe the details of the scope and major content of it. Then we propose the lifecycle of service system and give the definition of service engineering. Take the bicycle renting system as an example, this paper illustrate the main steps of service engineering.

Context-aware service is a new service mode that can provide appropriate service automatically to improve service level based on context information. In context-aware service, service is provided based on the current scene of customer. The scene is identified by context, and scene transition is caused by the change of context. Current research focuses on modeling service with context directly. However, the fact is that the vast majority of context change does not cause the change of scene, and monitoring context to determine the scene directly is inefficient, especially of the multi-context application. We use event which is defined by context as the motivation of scene change and propose an event driven model of context-aware service: EDM. A case of smart home service is discussed to show how to use EDM to do the requirement analysis, design, and implementation of context-aware service. Experimental results show that this approach can effectively reduce the number of scene determination in multi-context application.