Retrieval-Augmented Generation (RAG) significantly improves the factual accuracy and generation quality of large language models by incorporating external knowledge. However, in multilingual settings, RAG systems suffer from severe language preference. On the one hand, the retrieval stage is sensitive to the query language: semantically equivalent queries expressed in different languages often lead to substantially different retrieval results. On the other hand, when retrieved documents contain knowledge written in multiple languages, large language models tend to be influenced by surface-level language forms, rather than reasoning solely based on semantic relevance to the query.To address these challenges, we propose a unified optimization framework that explicitly disentangles multilingual RAG into language-controllable retrieval and language-agnostic reasoning. Our framework allows LLM to adaptively select retrieval languages while enforcing cross-lingual consistency during reasoning, thereby mitigating language bias without modifying existing retrievers or translators. Experimental results demonstrate that our approach effectively reduces language bias in multilingual RAG and consistently outperforms baselines across multiple multilingual benchmarks.

Stance detection is critical for understanding the underlying position or attitude expressed toward a topic. Large language models (LLMs) have demonstrated significant advancements across various natural language processing tasks including stance detection, however, their performance in stance detection is limited by biases and spurious correlations inherent due to their data-driven nature. Our statistical experiment reveals that LLMs are prone to generate biased stances due to sentiment-stance spurious correlations and preference towards certain individuals and topics. Furthermore, the results demonstrate a strong negative correlation between stance bias and stance detection performance, underscoring the importance of mitigating bias to enhance the utility of LLMs in stance detection. Therefore, in this paper, we propose a Counterfactual Augmented Calibration Network (FACTUAL), which a novel calibration network is devised to calibrate potential bias in the stance prediction of LLMs. Further, to address the challenge of effectively learning bias representations and the difficulty in the generalizability of debiasing, we construct counterfactual augmented data. This approach enhances the calibration network, facilitating the debiasing and out-of-domain generalization. Experimental results on in-target and zero-shot stance detection tasks show that the proposed FACTUAL can effectively mitigate biases of LLMs, achieving state-of-the-art results.

Efficient instruction tuning aims to enhance the ultimate performance of large language models (LLMs) trained on a given instruction dataset. Curriculum learning as a typical data organization strategy has shown preliminary effectiveness in instruction tuning. However, current curriculum tuning methods suffer from the curriculum rigidity, since they rely solely on static heuristic difficulty metrics. These methods fail to adapt to the evolving capabilities of models during training, resulting in a fixed and potentially sub-optimal learning trajectory. To address the issue, Competence-Aware Multi-Perspective cUrriculum inStruction tuning framework termed CAMPUS is proposed. CAMPUS offers several advantages: (1) Dynamic selection for sub-curriculum. (2) Competency-aware adjustment to the curriculum schedule. (3) Multiple difficulty-based scheduling. Extensive experiments prove the superior performance of CAMPUS, compared to other state-of-the-art baselines for efficient instruction tuning.

Reinforcement Learning from Human Feedback (RLHF) is effective for aligning Large Language Models (LLMs) with human preferences. However, RLHF’s complex process limits its ability to continually learn human feedback, making it impractical for real-world applications where the deployed model continuously receives feedback from users. The non-RL-based method, such as Direct Preference Optimization (DPO), is not primitively favorable for Continual Learning (CL). We observe that when combined with Experiment Relay (ER) for CL, DPO tends to significantly widen the gap in the probability of human-preferred and dispreferred responses. Consequently, this diminishes the diversity in model generation, potentially leading to model collapse. To overcome the above challenges, we propose the Continual Optimal Policy Regularization (COPR), a novel non-RL offline method to convert the historical optimal policies into optimization constraints when continually learning new preferences. We first derive a moderate reward function from the pairwise ranking loss and then use the moderate reward to calculate a new sampling distribution to construct novel learning objectives and constraints. We also provide formal proof of the learnability of COPR. The experimental results show that COPR outperforms strong CL baselines on our proposed benchmark, in terms of reward-based, GPT-4 evaluations and human assessment.

Graph-theoretic problems arise in real-world applications like logistics, communication networks, and traffic optimization. These problems are often complex, noisy, and irregular, posing challenges for traditional algorithms. Large language models offer potential solutions but face several challenges, including limited accuracy, input length constraints, and suboptimal algorithm selection. To address these challenges, we propose MA-GTS(Multi-Agent Graph Theory Solver), a multi-agent framework that decomposes these complex problems through agent collaboration. MA-GTS maps the implicitly expressed text-based graph data into clear, structured graph representations and dynamically selects the most suitable algorithm based on problem constraints and graph structure scale. We validate MA-GTS using the G-REAL dataset, a real-world-inspired graph theory dataset we created. Experimental results show that MA-GTS outperforms state-of-the-art methods in cost-effectiveness, accuracy, and scalability, achieving strong results on multiple benchmarks (G-REAL 93.6%, GraCoRe 96.9% ,NLGraph 98.4%) with robust performance on both closed- and open-source models.

Understanding the internal origins of capabilities in large language models (LLMs) is crucial for interpretability and efficient adaptation. However, the emergence of specific capabilities remains poorly understood, as most existing approaches rely on external signals (e.g., performance shifts or gradient similarities) with limited structural grounding. To address these issues, this paper proposes a lightweight and highly interpretable approach that links LLM capabilities to internal components by identifying correspondences at the level of attention heads. Specifically, we first define five fundamental capabilities, namely Mathematical Reasoning, Reading Comprehension, Commonsense Reasoning, Scientific Reasoning, and Professional Expertise, and employ probing techniques to detect the attention heads most predictive of each, thereby establishing capability–head mappings. For targeted instruction tuning, complex tasks are decomposed into these fundamental capabilities, and training data are selected accordingly. Experiments on LLaMA3.1-8B and Qwen2.5-7B show over 70% discrimination accuracy in identifying capabilities. On MMLU and BBH, our method improves accuracy by 1 to 1.5 points over the gradient-based method LESS and by 5 to 6 points over other intermediate-state baselines.

Large language models (LLMs) have made significant breakthroughs in extracting useful information from conversation history to enhance the response in long-term conversations. Summarizing useful information from historical conversations has achieved remarkable performance, which, however, may introduce irrelevant or redundant information, making it difficult to flexibly choose and integrate key information from different sessions during memory retrieval. To address this issue, we propose a Fragment-then-Compose framework, a novel memory utilization approach for long-term open-domain conversation, called FraCom. To be specific, inspired by the concept of proposition representation from Cognitive Psychology, we first represent the conversation history as a series of predicates plus arguments for propositional representation to preserve key information useful for memory (“Fragment”). Then, we compose propositional graphs for the conversation history based on the connection between shared arguments (“Compose”). During retrieval, we retrieve relevant propositions from the graph based on arguments from the current query. This essentially allows for flexible and effective utilization of related information in long-term memory for better response generation towards a query. Experimental results on four long-term open-domain conversation datasets demonstrate the effectiveness of our FraCom in memory utilization and its ability to enhance response generation for LLMs.

Supervised fine-tuning (SFT) has enabled large language models (LLMs) to exhibit promising performance on various tasks. However, this fine-tuning process only compares current predictions and labels on each sample, yet fails to perceive and understand its error outputs from different degrees, which may potentially produce a large percentage of serious errors. This poses a problem for aspect-based sentiment analysis (ABSA) in that these serious errors bring a greater negative impact than acceptable ones. Humans tend to compare mistakes to understand the varying degrees of mistakes, thus avoiding major bad decisions. Inspired by this, we propose a simple yet effective framework that could perceive and understand the degree of different errors by learning from comparative error pairs. It utilizes the SFT model to yield multiple outputs on each sample and selects acceptable and severe errors based on the acceptable scores. Together with the labels, we construct two comparative error pairs and exploit their calibration losses to optimize parameters. We conduct comprehensive experiments on ABSA datasets to demonstrate the effectiveness of our framework over baselines.

With the growing deployment of pre-trained models like Transformers on cloud platforms, privacy concerns about model parameters and inference data are intensifying. Existing Privacy-Preserving Transformer Inference (PPTI) frameworks face the “impossible trinity” of balancing privacy, efficiency, and performance: Secure Multi-Party Computation (SMPC)-based approaches ensure strong privacy but suffer from high computational overhead and performance losses; Conversely, permutation-based methods achieve near-plaintext efficiency and accuracy but compromise privacy by exposing sensitive model parameters and intermediate results. Bridging this gap with a single approach presents substantial challenges, motivating the introduction of CENTAUR, a groundbreaking PPTI framework that seamlessly integrates random permutations and SMPC to address the “impossible trinity”. By designing efficient PPTI algorithms tailored to the structural properties of Transformer models, CENTAUR achieves an unprecedented balance among privacy, efficiency, and performance. Our experiments demonstrate CENTAUR’s ability to resist diverse data reconstruction attacks, achieve plaintext-level inference accuracy, and boost inference speed by 5.0~30.4 times, unlocking new possibilities for secure and efficient AI deployment.

Large language models have recently become a new learning paradigm and led to state-of-the-art performance across a range of tasks. As explosive open-source pre-trained models are available, it is worth investigating how to better utilize existing models. We propose a simple yet effective method, Incr-Pretrain, for incrementally pre-training language models from smaller well-trained source models. Different layer-wise transfer strategies were introduced for model augmentation including parameter copying, initial value padding, and model distillation. Experiments on multiple zero-shot learning tasks demonstrate satisfying inference performance upon transferring and promising training efficiency during continuing pre-training. Compared to training from scratch, Incr-Pretrain can save up to half the training time to get a similar testing loss.