Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Hai ZhaoSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works182 from public data
- Multimodal Chain-of-Thought Reasoning in Language Models966
This work proposes Multimodal-CoT that incorporates language (text) and vision (images) modalities into a two-stage framework that separates rationale generation and answer inference, and indicates that it offers the advantages of mitigating hallucination and enhancing convergence speed.
- Is ChatGPT a General-Purpose Natural Language Processing Task Solver?894
It is found that ChatGPT performs well on many tasks favoring reasoning capabilities while it still faces challenges when solving specific tasks such as sequence tagging, and with extensive empirical studies, both the effectiveness and limitations of the current version of ChatG PT are demonstrated.
- Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs586
This paper introduces novel efficiency metrics from both outcome and process perspectives to evaluate the rational use of computational resources by o1-like models, and proposes strategies to mitigate overthinking, streamlining reasoning processes without compromising accuracy.
- Semantics-Aware BERT for Language Understanding407
This work proposes to incorporate explicit contextual semantics from pre-trained semantic role labeling, and introduces an improved language representation model, Semantics-aware BERT (SemBERT), which is capable of explicitly absorbing contextual semantics over a BERT backbone.
- DeepMath-103K: A Large-Scale, Challenging, Decontaminated, and Verifiable Mathematical Dataset for Advancing Reasoning282
DeepMath-103K is introduced, a large-scale mathematical dataset designed with high difficulty, rigorous decontamination against numerous benchmarks, and verifiable answers for rule-based RL reward that fosters the development of generalizable and advancing reasoning.
- Retrospective Reader for Machine Reading Comprehension241
A retrospective reader (Retro-Reader) that integrates two stages of reading and verification strategies: 1) sketchy reading that briefly investigates the overall interactions of passage and question, and yields an initial judgment; 2) intensive reading that verifies the answer and gives the final prediction.
- 232
- SG-Net: Syntax-Guided Machine Reading Comprehension199
This work uses syntax to guide the text modeling by incorporating explicit syntactic constraints into attention mechanism for better linguistically motivated word representations and shows that the proposed SG-Net design helps achieve substantial performance improvement over strong baselines.
- Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMs172
A decoding strategy with thought switching penalty TIP that discourages premature transitions between thoughts, encouraging deeper exploration of each reasoning path and contributes to understanding reasoning inefficiencies in o1-like LLMs and offers a practical solution to enhance their problem-solving capabilities.
- DCMN+: Dual Co-Matching Network for Multi-Choice Reading Comprehension141
This work proposes dual co-matching network (DCMN) which models the relationship among passage, question and answer options bidirectionally and integrates two reading strategies into the model, inspired by how humans solve multi-choice questions.
- 135
- Igniting Language Intelligence: The Hitchhiker’s Guide from Chain-of-Thought Reasoning to Language Agents126
A thorough discourse is orchestrated, penetrating vital research dimensions, encompassing the foundational mechanics of CoT techniques, with a focus on elucidating the circumstances and justification behind its efficacy; the paradigm shift in CoT; and the burgeoning of language agents fortified by CoT approaches.
- Neural Machine Translation with Universal Visual Representation124
A universal visual representation learned over the monolingual corpora with image annotations is presented, which overcomes the lack of large-scale bilingual sentence-image pairs, thereby extending image applicability in NMT.
- Exploring Human-Like Translation Strategy with Large Language Models118
This work proposes the MAPS framework, which stands for Multi-Aspect Prompting and Selection, and enables LLMs first to analyze the given source sentence and induce three aspects of translation-related knowledge to guide the final translation process.
- Dependency or Span, End-to-End Uniform Semantic Role Labeling102
This paper presents an end-to-end model for both dependency and span SRL with a unified argument representation to deal with two different types of argument annotations in a uniform fashion and jointly predict all predicates and arguments.
- Risks of AI scientists: prioritizing safeguarding over autonomy98
The authors examine the vulnerabilities of AI scientists, propose a risk taxonomy based on user intent and impact domains, and develop a triadic safeguarding framework emphasizing human regulation, agent alignment, and environmental feedback understanding.
- Topic-Aware Multi-turn Dialogue Modeling97
This paper presents a novel topic-aware solution for multi-turn dialogue modeling, which segments and extracts topic- aware utterances in an unsupervised way, so that the resulted model is capable of capturing salient topic shift at discourse-level in need and thus effectively track topic flow during multi- turn conversation.
- Flooding spread of manipulated knowledge in LLM-based multi-agent communities95
This paper proposes a novel two-stage attack method involving Persuasiveness Injection and Manipulated Knowledge Injection to systematically explore the potential for manipulated knowledge to spread without explicit prompt manipulation and demonstrates that it can successfully induce LLM-based agents to spread both counterfactual and toxic knowledge without degrading their foundational capabilities during agent communication.
- A Unified Syntax-aware Framework for Semantic Role Labeling91
This work extends existing models and proposes a unified framework to comprehensively explore the role of syntax for SRL task, and confirms that high-quality syntactic parse could still effectively enhance syntactically-driven SRL.
- 90
- TrojanRAG: Retrieval-Augmented Generation Can Be Backdoor Driver in Large Language Models87
This paper proposes TrojanRAG, which employs a joint backdoor attack in the Retrieval-Augmented Generation, thereby manipulating LLMs in universal attack scenarios and shows that TrojanRAG exhibits versatility threats while maintaining retrieval capabilities on normal queries.
- Self-Prompting Large Language Models for Zero-Shot Open-Domain QA74
A Self-Prompting framework to explicitly utilize the massive knowledge encoded in the parameters of LLMs and their strong instruction understanding abilities and even achieves comparable performance with various customized fine-tuned models on full training data.
- Data-dependent Gaussian Prior Objective for Language Generation70
Experimental results show that the proposed data-dependent Gaussian prior objective (D2GPo) can effectively make use of more detailed prior in the data and significantly improve the performance of typical language generation tasks, including supervised and unsupervised machine translation, text summarization, storytelling, and image caption.
- Filling the Gap of Utterance-aware and Speaker-aware Representation for Multi-turn Dialogue57
A novel model is proposed to fill the gap in retrieval-based multi-turn dialogue modeling by modeling the effective utterance-aware and speaker-aware representations entailed in a dialogue history by decouple the contextualized word representations by masking mechanisms in Transformer-based PrLM.
- SAFELearning: Secure Aggregation in Federated Learning With Backdoor Detectability55
A new federated learning technique SAFELearning is proposed to support backdoor detection for secure aggregation and is robust against malicious and faulty participants, whilst being more efficient than the state-of-art secure aggregation protocol in terms of both communication and computation costs.
- Text Compression-aided Transformer Encoding54
This paper proposes explicit and implicit text compression approaches to enhance the Transformer encoding and evaluates models using this approach on several typical downstream tasks that rely on the encoding heavily, and concludes that text compression helps the encoders to learn better language representations.
- LIMIT-BERT : Linguistics Informed Multi-Task BERT48
LIMIT-BERT outperforms the strong baseline Whole Word Masking BERT on both dependency and constituent syntactic/semantic parsing, GLUE benchmark, and SNLI task and is able to release a well pre-trained model for multi-purpose of natural language processing tasks once for all.
- Dialogue Graph Modeling for Conversational Machine Reading47
A dialogue graph modeling framework to improve the understanding and reasoning ability of machine on CMR task and learns explicitly and extracts the discourse relation among rule texts as well as the extra knowledge of scenario.
- 45
- 45
- 42
- 41
- Enhanced Speaker-Aware Multi-Party Multi-Turn Dialogue Comprehension40
The Enhanced Speaker-Aware method (ESA) helps achieve state-of-the-art performance on the Molweni dataset, as well as significant improvements on the FriendsQA dataset, and analysis shows that the method makes steady improvements on stronger backbones.
- Moon IME: Neural-based Chinese Pinyin Aided Input Method with Customizable Association39
The Moon IME is presented, a pinyin IME that integrates the attention-based neural machine translation (NMT) model and Information Retrieval (IR) to offer amusive and customizable association ability.
- Universal Multimodal Representation for Language Understanding38
Analysis indicates that the visual signals enrich textual representations of content words, provide fine-grained grounding information about the relationship between concepts and events, and potentially conduce to disambiguation.
- 37
- Effective Subword Segmentation for Text Comprehension36
A series of unsupervised segmentation methods for subword acquisition and different subword-augmented strategies for text understanding are surveyed, showing that sub word-AUgmented embedding significantly improves the authors' baselines in various types of text understanding tasks on both English and Chinese benchmarks.
- Structural Pre-training for Dialogue Comprehension34
SPIDER, Structural PretraIned DialoguE Reader, is presented, to capture dialogue exclusive features and proposes two training objectives in addition to the original LM objectives, which regularizes the model to improve the factual correctness of summarized subject-verb-object triplets.
- When SMILES Smiles, Practicality Judgment and Yield Prediction of Chemical Reaction via Deep Chemical Language Processing32
This work presents a symbol only model to generally predict the yield of organic synthesis reaction without considering complex quantum physical modeling or chemistry knowledge, which achieves a high accuracy and low modeling cost.
- Explicit Sentence Compression for Neural Machine Translation32
This paper proposed an explicit sentence compression method to enhance the source sentence representation for NMT, and proposes three ways, including backbone source-side fusion, target- side fusion, and both-side fused, to integrate the compressed sentence into NMT.
- 32
- Structured Chemistry Reasoning with Large Language Models31
Testing across four chemistry areas -- quantum chemistry, mechanics, physical chemistry, and kinetics -- StructChem substantially enhances GPT-4's performance, with up to 30\% peak improvement, and underscores the unique difficulties of precise grounded reasoning in science with LLMs.
- Sentence-aware Contrastive Learning for Open-Domain Passage Retrieval30
This work introduces an in-passage negative sampling strategy to encourage a diverse generation of sentence representations within the same passage, and presents a refined model on the basis of a smaller granularity, contextual sentences, to alleviate the concerned conflicts.
- 30
- Open Vocabulary Learning for Neural Chinese Pinyin IME28
The proposed neural P2C conversion model augmented by an online updated vocabulary with a sampling mechanism to support open vocabulary learning during IME working outperforms commercial IMEs and state-of-the-art traditional models on standard corpus and true inputting history dataset in terms of multiple metrics.
- 27
- 27
- Structural Characterization for Dialogue Disentanglement26
This work specially takes structure factors into account and design a novel model for dialogue disentangling that achieves new state-of-the-art on the Ubuntu IRC benchmark dataset and contributes to dialogue-related comprehension.
- Multi-Turn Dialogue Reading Comprehension With Pivot Turns and Knowledge25
This work proposes a pivot-oriented deep selection model (PoDS) on top of the Transformer-based language models for dialogue comprehension by extracting substantially important turns as pivot utterances and utilizing external knowledge to enhance the representation of context.
- 24
- Task-specific Objectives of Pre-trained Language Models for Dialogue Adaptation23
A Dialogue-Adaptive Pre-training Objective (DAPO) based on some important qualities for assessing dialogues which are usually ignored by general LM pre-training objectives is designed and Experimental results show that models with DAPO surpass those with general LMPre- training objectives and other strong baselines on downstream DrNLP tasks.
- Joint Learning of23
This system predicts the part-of-speech tag and dependency tree jointly and achieves a macro-average of 68.31% LAS F1 score, with an improvement of 2.51% compared with the UDPipe.
- 22
- 21
- I Know What You Want: Semantic Learning for Text Comprehension21
This paper makes the first attempt to let SRL enhance text comprehension and inference through specifying verbal arguments and their corresponding semantic roles, and shows that the salient labels can be conveniently added to existing models and significantly improve deep learning models in challenging text comprehension tasks.
- DeepTheorem: Advancing LLM Reasoning for Theorem Proving Through Natural Language and Reinforcement Learning19
Extensive experimental analyses demonstrate DeepTheorem significantly improves LLM theorem-proving performance compared to existing datasets and supervised fine-tuning protocols, achieving state-of-the-art accuracy and reasoning quality.
- Effective Character-Augmented Word Embedding for Machine Reading Comprehension19
A character-augmented reader which attends character-level representation to augment word embedding with a short list to improve word representations, especially for rare words is proposed.
- Rethinking Textual Adversarial Defense for Pre-Trained Language Models18
A universal defense framework is designed, which is among the first to perform textual adversarial defense without knowing the specific attack and achieves comparable or even higher after-attack accuracy with other specific defenses, while preserving higher original accuracy at the same time.
- 18
- SJTU-NICT at MRP 2019: Multi-Task Learning for End-to-End Uniform Semantic Graph Parsing17
This paper describes SJTU-NICT’s system for participating in the shared task on Cross-Framework Meaning Representation Parsing (MRP) at the 2019 Conference for Computational Language Learning (CoNLL) and introduces multi-task learning for multiple objectives within the same framework.
- Dancing with Critiques: Enhancing LLM Reasoning with Stepwise Natural Language Self-Critique16
This paper proposes a novel inference-time scaling approach -- stepwise natural language self-critique (PANEL), which employs self-generated natural language critiques as feedback to guide the step-level search process, and demonstrates that PANEL significantly enhances reasoning performance, outperforming traditional scalar reward-based methods.
- Dialogue-adaptive language model pre-training from quality estimation☆15
Experimental results on widely used open-domain response selection and quality estimation benchmarks show that DAPO significantly improves the baseline models and achieves state-of-the-art performance on the MuTual leaderboard, verifying the effectiveness of estimating quality evaluation factors into pre-training.
- VeriOS: Query-Driven Proactive Human-Agent-GUI Interaction for Trustworthy OS Agents14
VeriOS-Agent is introduced, a trustworthy OS agent trained with a three-stage learning paradigm that falicitate the decoupling and utilization of meta-knowledge by supervised fine-tuning and group relative policy optimization that significantly improves performance in untrustworthy scenarios while maintaining comparable performance in trustworthy scenarios.
- Plan-over-Graph: Towards Parallelable LLM Agent Schedule14
A novel paradigm, plan-over-graph, in which the model first decomposes a real-life textual task into executable subtasks and constructs an abstract task graph, which naturally supports parallel execution, demonstrating global efficiency.
- MEGen: Generative Backdoor into Large Language Models via Model Editing14
This paper proposes an editing-based generative backdoor, named MEGen, aiming to expand the backdoor to generative tasks in a unified format of any text-to any text, leading to natural generations with a specific intention.
- Mitigating Misleading Chain-of-Thought Reasoning with Selective Filtering14
A novel approach called the selective filtering reasoner (SelF-Reasoner) that assesses the entailment relationship between the question and the candidate reasoning chain and improves the fine-tuned T5 baseline consistently over the ScienceQA, ECQA, and LastLetter tasks.
- Thoughts Are All Over the Place: On the Underthinking of Long Reasoning Models13
A decoding strategy with thought switching penalty (T IP) that discourages premature transitions between thoughts, encouraging deeper exploration of each reasoning path and improves accuracy across challenging datasets without requiring model fine-tuning is proposed.
- From Multimodal LLM to Human-level AI: Modality, Instruction, Reasoning and Beyond13
This tutorial aims to deliver a comprehensive review of cutting-edge research in MLLMs, focusing on three key areas: MLLM architecture design, instructional learning, and multimodal reasoning of MLLMs.
- 13
- Atomic-to-Compositional Generalization for Mobile Agents with A New Benchmark and Scheduling System12
This work introduces UI-NEXUS, a comprehensive benchmark designed to evaluate mobile agents on three categories of compositional operations: Simple Concatenation, Context Transition, and Deep Dive, and proposes AGENT-NEXUS, a lightweight and efficient scheduling system to tackle compositional mobile tasks.
- 12
- Fact-Driven Logical Reasoning for Machine Reading Comprehension12
A general formalism of knowledge units is proposed by extracting backbone constituents of the sentence, such as the subject-verb-object formed ``facts'' and constructing a supergraph on top of the fact units, allowing for the benefit of sentence-level (relations among fact groups) and entity-level interactions.
- 12
- 12
- Reference Knowledgeable Network for Machine Reading Comprehension12
A novel reference-based knowledge enhancement model called RekNet, which simulates human reading strategies to refine critical information from the passage and quote explicit knowledge in necessity, obtaining consistent and remarkable performance improvement with observable statistical significance level over strong baselines.
- 12
- 12
- Smoothing Grounding and Reasoning for MLLM-Powered GUI Agents with Query-Oriented Pivot Tasks11
This work proposes a query-oriented pivot approach called query inference, which serves as a bridge between GUI grounding and reasoning, and shows that query inference outperforms previous grounding techniques under the same training data scale.
- Quick on the Uptake: Eliciting Implicit Intents from Human Demonstrations for Personalized Mobile-Use Agents11
IFRAgent is proposed, a framework built upon Intention Flow Recognition from human demonstrations that leverages a SOP extractor combined with retrieval-augmented generation and a query rewriter trained through knowledge distillation to generate personalized queries and SOPs from a raw ambiguous query, enhancing the alignment between mobile-use agents and human intent.
- Gracefully Filtering Backdoor Samples for Generative Large Language Models without Retraining11
This work proposes Gradient Clustering in the Frequency Space for Backdoor Sample Filtering (GraCeFul), which leverages sample-wise gradients in the frequency space to effectively identify backdoor samples without requiring retraining LLMs.
- SJTU-NLP at SemEval-2018 Task 9: Neural Hypernym Discovery with Term Embeddings11
A hypernym discovery system for the SemEval-2018 Task 9, which aims to discover the best (set of) candidate hypernyms for input concepts or entities, given the search space of a pre-defined vocabulary, is described.
- ColorBench: Benchmarking Mobile Agents with Graph-Structured Framework for Complex Long-Horizon Tasks10
A novel graph-structured benchmarking framework that achieves static simulation of dynamic behaviors by modeling the finite states observed during real-device interactions, and develops ColorBench, a benchmark focused on complex long-horizon tasks.
- EVA: Evolving Semantic Adversaries for Red-Teaming GUI Agents Against Environmental Injection Attacks10
This work introduces EVA, an evolutionary framework that evolves adversarial payloads exclusively within the semantic dimension, and employs a discovery-deployment framework to mine linguistic vulnerability patterns and distill them into generalizable rules.
- 10
- Modeling Hierarchical Reasoning Chains by Linking Discourse Units and Key Phrases for Reading Comprehension10
A holistic graph network (HGN) that deals with context at both discourse-level and word-level as the basis for logical reasoning to provide a more fine-grained relation extraction and is modeled by a hierarchical interaction mechanism to improve the interpretation of MRC systems.
- 10
- SynGhost: Invisible and Universal Task-agnostic Backdoor Attack via Syntactic Transfer9
An entropy-based poisoning filter that mitigates task-agnostic backdoor attacks due to vulnerabilities in data and training mechanisms and injects multiple syntactic backdoors into the pre-training space through corpus poisoning, while preserving the PLM's pre-training capabilities.
- 9
- 9
- 9
- 9
- 8
- 8
- Tracing Origins: Coreference-aware Machine Reading Comprehension8
This paper imitate the human reading process in connecting the anaphoric expressions and explicitly leverage the coreference information of the entities to enhance the word embeddings from the pre-trained language model, in order to highlight thecoreference mentions of the entity that must be identified for coreference-intensive question answering in QUOREF.
- 8
- 8
- Revisiting Data Auditing in Large Vision-Language Models7
This study presents a systematic view of the limits and opportunities of MI for VLMs, providing guidance for future efforts in trustworthy data auditing.
- Smoothing Dialogue States for Open Conversational Machine Reading7
This work proposes an effective gating strategy by smoothing the two dialogue states in only one decoder and bridge decision making and question generation to provide a richer dialogue state reference and achieves new state-of-the-art results.
- Agent-Dice: Disentangling Knowledge Updates via Geometric Consensus for Agent Continual Learning6
This work argues that the stability-plasticity dilemma fundamentally arises from the failure to explicitly distinguish between common knowledge shared across tasks and conflicting knowledge introduced by task-specific interference, and proposes Agent-Dice, a parameter fusion framework based on directional consensus evaluation.
- Retrospective Progress-Aware Self-Refinement for LLM Agent Training6
Policy optimization combines self-generated progress differences, online-retrospective alignment, and format rewards with environment feedback, requiring neither a separate process reward model nor ongoing teacher annotation.
- Say One Thing, Do Another? Diagnosing Reasoning-Execution Gaps in VLM-Powered Mobile-Use Agents6
A new evaluation framework is introduced to diagnose reasoning-execution gaps and support the development of more trustworthy mobile-use agents by combining GTA with the standard Exact Match metric, which measures whether the action implied by a CoT matches the ground-truth action.
- Improved Secure Deep Neural Network Inference Offloading with Privacy-Preserving Scalar Product Evaluation for Edge Computing6
This paper addresses the challenge of efficiently executing inference tasks without hampering the real-time operation of IoT applications by utilizing low-latency IoT-to-edge communication links for computational efficiency and offers improved trade-offs among privacy, efficiency, and utility.
- 6
- Generalizable and Adaptive Continual Learning Framework for AI-Generated Image Detection5
A novel three-stage domain continual learning framework designed for continuous adaptation to evolving generative models, and establishes a comprehensive benchmark of 27 generative models, including GANs, deepfakes, and diffusion models, chronologically structured up to August 2024 to simulate real-world scenarios.
- 5
- 5
- Generalizable Chain-of-Thought Prompting in Mixed-task Scenarios with Large Language Models5
GeM-CoT is proposed, a Generalizable CoT prompting mechanism in Mixed-task scenarios where the type of input questions is unknown, which enjoys superior generalization capabilities and remarkable performances on 10 public reasoning tasks and 23 BBH tasks.
- 5
- Open Named Entity Modeling From Embedding Distribution5
The initial visualization of monolingual word embeddings indicates named entities tend to gather together despite of named entity types and language difference, which enable us to model all named entities using a specific geometric structure inside embedding space, namely, the named entity hypersphere.
- 5
- 4
- Chain-of-Trigger: An Agentic Backdoor that Paradoxically Enhances Agentic Robustness4
This work proposes the Chain-of-Trigger Backdoor (CoTri), a multi-step backdoor attack designed for long-horizon agentic control that achieves stable, multi-step control within agents, improving their inherent robustness and task capabilities.
- 4
- GLaPE: Gold Label-agnostic Prompt Evaluation and Optimization for Large Language Model4
Experimental results on 8 widely-recognized reasoning tasks demonstrate that GLaPE can produce more effective prompts, achieving performance comparable to those derived from manually annotated gold labels.
- 4
- Towards End-to-End Open Conversational Machine Reading4
This work model OR-CMR as a unified text- to-text task in a fully end-to-end style and shows the effectiveness of the proposed end-To-end framework on both sub-tasks by a large margin, achieving new state-of-the-art results.
- Channel-Aware Decoupling Network for Multiturn Dialog Comprehension4
This work proposes compositional learning for holistic interaction across the utterances beyond the sequential contextualization from PrLMs, in order to capture the utterance-aware and speaker-aware representations entailed in a dialog history.
- 4
- 4
- SE-GA: Memory-Augmented Self-Evolution for GUI Agents3
The Self-Evolving GUI Agent (SE-GA) is proposed, a novel framework that integrates hierarchical memory structures with an iterative self-improvement mechanism that facilitates long-term planning by dynamically retrieving episodic, semantic, and experiential memories during inference.
- Watch Out Your Album! On the Inadvertent Privacy Memorization in Multi-Modal Large Language Models3
This paper randomly generates task-irrelevant watermarks into VQA fine-tuning images at varying probabilities and proposes a novel probing framework to determine whether MLLMs have inadvertently encoded private content, thus causing inadvertent memorization.
- 3
- See, Think, Act: Teaching Multimodal Agents to Effectively Interact with GUI by Identifying Toggles3
A state control benchmark with binary toggle instructions derived from public datasets is constructed and proposed State-aware Reasoning (StaR), a multimodal reasoning method that enables agents to perceive the current toggle state, infer the desired state from the instruction, and act accordingly.
- 3
- When Disagreements Elicit Robustness: Investigating Self-Repair Capabilities under LLM Multi-Agent Disagreements3
This paper designs four comprehensive metrics to investigate the robustness of MASs when facing mild or task-critical knowledge conflicts, and finds that the self-repairing capability of MASs has intrinsic limits, and all findings hold consistently across various factors.
- On the Adaptive Psychological Persuasion of Large Language Models3
This paper proposes an adaptive framework based on direct preference optimization that trains LLMs to autonomously select optimal strategies by leveraging persuasion results from strategy-specific responses as preference pairs, and demonstrates that the proposed adaptive psychological persuasion method effectively enables persuader LLMs to select optimal strategies.
- 3
- 3
- How Deep is Love in LLMs' Hearts? Exploring Semantic Size in Human-like Cognition2
The findings reveal that multi-modal training is crucial for LLMs to achieve more human-like understanding, suggesting that real-world, multi-modal experiences are similarly vital for human cognitive development.
- 2
- Look Before You Leap: Enhancing Attention and Vigilance Regarding Harmful Content with GuidelineLLM2
This work proposes a novel defensive paradigm called GuidelineLLM, which assists LLMs in recognizing queries that may have harmful content and can significantly reduce the attack success rate (ASR) against LLM while maintaining the usefulness of LLM in handling benign queries.
- 2
- 2
- Span Fine-tuning for Pre-trained Language Models2
A novel span fine-tuning method for PrLMs is presented, which facilitates the span setting to be adaptively determined by specific downstream tasks during the fine- Tuning phase.
- 2
- 2
- 2
- 2
- 2
- Neural-based Pinyin-to-Character Conversion with Adaptive Vocabulary.2
A neural P2C conversion model augmented by a large online updating vocabulary with a target vocabulary sampling mechanism is proposed that reduces the decoding time on CPUs up to 50$\%$ on pinyin-to-character tasks at the same or only negligible change in conversion accuracy.
- WHU-NERCMS at TRECVID2014: Instance Search Task.2
The work at the automatic instance search task of TRECVID 2014 is divided into two parts: first part is object retrieval based on BOW, and secondly, several optimization strategies are adopted to improve the initial results.
- 1
- Beyond Elicitation: Provision-Based Prompt Optimization for Knowledge-Intensive Tasks1
Evaluation on 15 knowledge-intensive benchmarks from various domains demonstrates KPPO’s superiority over elicitation-based methods, with an average improvement of ~6% over baselines while achieving comparable or lower token consumption.
- 1
- 1
- Look Before You Leap: Enhance Attention and Vigilance Regarding Harmful Content with GuidelineLLM1
A novel defensive paradigm called GuidelineLLM is proposed, which assists LLMs in recognizing queries that may have harmful content and can significantly reduce the attack success rate (ASR) against LLM while maintaining the usefulness of LLM in handling benign queries.
- 1
- 1
- Embedding Trajectory for Out-of-Distribution Detection in Mathematical Reasoning1
A trajectory-based method TV score is proposed, which uses trajectory volatility for OOD detection in mathematical reasoning and outperforms all traditional algorithms on GLMs under mathematical reasoning scenarios and can be extended to more applications with high-density features in output spaces, such as multiple-choice questions.
- NSmark: Null Space Based Black-box Watermarking Defense Framework for Language Models1
This paper analyzes and extends the attack scenarios of LFEA to the commonly employed black-box settings for PLMs by considering Last-Layer outputs (dubbed LL-LFEA), and discovers that the null space of the output matrix remains invariant against LL-LFEA attacks.
- 1
- EventBERT: Incorporating Event-Based Semantics for Natural Language Understanding1
This work proposes EventBERT, an event-based semantic representation model that takes BERT as the backbone and refines with event- based structural semantics in terms of graph convolution networks, and shows that the proposed model consistently outperforms the baseline model.
- 1
- 1
- 1
- 1
- 1
- Multi-choice Dialogue-Based Reading Comprehension with Knowledge and Key Turns1
This work makes the first attempt to tackle two challenges of multi-choice machine reading comprehension by extracting substantially important turns and utilizing external knowledge to enhance the representation of context.
- 1
- OS-SPEAR: A Toolkit for the Safety, Performance,Efficiency, and Robustness Analysis of OS Agents–
The empirical results reveal critical insights into the current landscape: notably, a prevalent trade-off between efficiency and safety or robustness, the performance superiority of specialized agents over general-purpose models, and varying robustness vulnerabilities across different modalities.
- Post-quantum Cryptography in the Wild: Assessing the Readiness of Open-Source Ecosystems–
PQC Scanner is a two-stage pipeline in which crypto detection identifies packages with cryptographic functionality, and PQC detection produces a binary support label and source-derived structured readiness record at the repository level using commit history, source analysis, and agent post-verification.
- –
- An LLM-based Quantitative Framework for Evaluating High-Stealthy Backdoor Risks in OSS Supply Chains–
- –
- Beyond a Single Judge: The Evidence-Grounded, Social-Weighted Persona Panel for Generative UI Evaluation–
The Evidence-Grounded, Social-Weighted Persona Panel is proposed, a three-stage GenUI evaluation method in which a panel of psychologically diverse, evidence-grounded personas independently rates a screenshot, exchanges opinions under a trait-derived, semantically-gated bounded-confidence mechanism, and is aggregated via Delphi-inspired social weighting into a single judgment.
- –
- Mobile-Aptus: Confidence-Driven Proactive and Robust Interaction in MLLM-Based Mobile-Using Agents–
This work proposes a universal confidence integration framework that enables confidence-driven proactive and robust interaction in MLLM-based mobile-using agents and finds that the salient human interventions can be effectively replaced by a multi-agent system to avoid task interruption.
- –
- Communication Policy Evolution for Proactive LLM Agents–
This paper formalizes Communication Policy, establishes textual and UI-based policies, and then evaluates communication policies across diverse environments, personas, and model combinations, identifying communication behavior as a critical yet underexplored design dimension for LLM agents.
- –
- ColorEcosystem: Powering Personalized, Standardized, and Trustworthy Agentic Service in massive-agent Ecosystem–
Through the analysis of challenges, transitional forms, and practical considerations, the ColorEcosystem is poised to power personalized, standardized, and trustworthy agentic service across massive-agent ecosystems.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-12. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.