Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks38 from public data
- Triggerless Backdoor Attack for NLP Tasks with Clean Labels109
A new strategy to perform textual backdoor attack which does not require an external trigger and the poisoned samples are correctly labeled is proposed, which marks the first step towards developing triggerless attacking strategies in NLP.
- Judgment Prediction via Injecting Legal Knowledge into Neural Networks79
This paper proposes to represent declarative legal knowledge as a set of first-order logic rules and integrate these logic rules into a co-attention network-based model explicitly, which enhances neural networks with explicit logical reason capabilities and makes the model more interpretable.
- Hybrid computation offloading for smart home automation in mobile cloud computing57
This work model the waiting time spending in the cloudlet as a M/M/m/∞ queue and proposes a hybrid computation offloading algorithm for home automation applications to minimize the total energy consumption of the mobile devices within a given constant deadline.
- Fast-Slow Thinking GRPO for Large Vision-Language Model Reasoning34
FAST-GRPO, a variant of GRPO that dynamically adapts reasoning depth based on question characteristics, is presented, establishing the feasibility of fast-slow thinking in LVLMs by investigating how response length and data distribution affect performance.
- A Survey of Direct Preference Optimization: Datasets, Theories, Variants, and Applications27
This work categorizes recent studies on DPO based on key research questions to provide a thorough understanding of DPO’s current landscape, and proposes several future research directions to offer insights on model alignment for the research community.
- Backdoor Attacks on Dense Retrieval via Public and Unintentional Triggers26
A novel scenario where the attackers aim to covertly disseminate targeted misinformation through a retrieval system through a retrieval system triggered by grammar errors in dense passage retrieval is introduced.
- Dependency Parsing as MRC-based Span-Span Prediction24
The proposed method addresses the fundamental problem that edges in a dependency tree should be constructed between subtrees and the MRC framework allows the method to retrieve missing spans in the span proposal stage, which leads to higher recall for eligible spans.
- SemGloVe: Semantic Co-Occurrences for GloVe From BERT24
SemGloVe is proposed, which distills semantic co-occurrences from BERT into static GloVe word embeddings and can define the co- Occurrence weights by directly considering the semantic distance between word pairs.
- Detecting and Mitigating Hallucination in Large Vision Language Models via Fine-Grained AI Feedback21
This work proposes a detect-then-rewrite pipeline to automatically construct preference dataset for hallucination mitigation training, and introduces a Hallucination Severity-Aware Direct Preference Optimization (HSA-DPO) which prioritizes the mitigation of critical hallucination in LVLMs by incorporating the severity of hallucinations into preference learning.
- 20
- 20
- 20
- Investigating Self-Attention Network for Chinese Word Segmentation19
This work investigates self-attention network (SAN) for Chinese word segmentation, making comparisons between BiLSTM-CRF models and the influence of contextualized character embeddings using BERT, and proposes a method for integrating word information into SAN segmentation.
- Retrieval-Augmented Mixture of LoRA Experts for Uploadable Machine Learning18
Retrieval-Augmented Mixture of LoRA Experts (RAMoLE), a framework that adaptively retrieves and composes multiple LoRAs based on input prompts, is proposed, highlighting its effectiveness and scalability.
- Exploiting Contrastive Learning and Numerical Evidence for Confusing Legal Judgment Prediction18
A moco-based supervised contrastive learning to learn distinguishable representations is proposed, and the best strategy to construct positive example pairs to benefit all three subtasks of LJP simultaneously is explored.
- T2I-FactualBench: Benchmarking the Factuality of Text-to-Image Models with Knowledge-Intensive Concepts16
This work presents T2I-FactualBench - the largest benchmark to date in terms of the number of concepts and prompts specifically designed to evaluate the factuality of knowledge-intensive concept generation, and introduces a multi-round visual question answering (VQA) based evaluation framework to assess the factuality of three-tiered knowledge-intensive text-to-image generation tasks.
- Evaluating Test-Time Scaling LLMs for Legal Reasoning: OpenAI o1, DeepSeek-R1, and Beyond15
This work presents the first systematic evaluation of 12 LLMs, including both reasoning-focused and general-purpose models, across 17 Chinese and English legal tasks spanning statutory and case-law traditions and develops Legal-R1, an open-source model specialized for the legal domain.
- Don't Just Fine-tune the Agent, Tune the Environment14
This work presents a paradigm shift from supervised fine-tuning on static trajectories to dynamic, environment-based exploration, paving the way for training more robust and data-efficient agents.
- SQLCritic: Correcting Text-to-SQL Generation via Clause-wise Critic9
A novel approach combining structured execution feedback with a trained critic agent that provides detailed, interpretable critiques is proposed, which effectively identifies and corrects both syntactic and semantic errors, enhancing accuracy and interpretability.
- 9
- 8
- 8
- V2P: Visual Attention Calibration for GUI Grounding via Background Suppression and Center Peaking5
V2P introduces a suppression attention mechanism that minimizes the model's focus on irrelevant regions to highlight the intended region, and applies a Fitts'Law-inspired approach by modeling GUI interactions as 2D Gaussian heatmaps where the weight gradually decreases from the center towards the edges.
- Self-Verification Dilemma: Experience-Driven Suppression of Overused Checking in LLM Reasoning4
This work proposes a novel, experience-driven test-time framework that reduces the overused verification in large Reasoning Models, and detects the activation of recheck behavior, consults an offline experience pool of past verification outcomes, and estimates whether a recheck is likely unnecessary via efficient retrieval.
- 2
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.