Reinforced Informativeness Optimization for Long-Form Retrieval-Augmented Generation

Long-form question answering (LFQA) presents unique challenges for large language models, requiring the synthesis of coherent, paragraph-length answers. While retrieval-augmented generation (RAG) systems have emerged as a promising solution, existing research struggles with key limitations: the scarcity of high-quality training data for long-form generation, the compounding risk of hallucination in extended outputs, and the absence of reliable evaluation metrics for factual completeness. In this paper, we propose RioRAG, a novel reinforcement learning (RL) framework that advances long-form RAG through reinforced informativeness optimization. Our approach introduces two fundamental innovations to address the core challenges. First, we develop an RL training paradigm of reinforced informativeness optimization that directly optimizes informativeness and effectively addresses the slow-thinking deficit in conventional RAG systems, bypassing the need for expensive supervised data. Second, we propose a nugget-centric hierarchical reward modeling approach that enables precise assessment of long-form answers through a three-stage process: extracting the nugget from every source webpage, constructing a nugget claim checklist, and computing rewards based on factual alignment. Extensive experiments on two LFQA benchmarks LongFact and RAGChecker demonstrate the effectiveness of the proposed method. Our codes are available at https://github.com/RUCAIBox/RioRAG.

C-LoRA: Contextual Low-Rank Adaptation for Uncertainty Estimation in Large Language Models

Low-Rank Adaptation (LoRA) offers a cost-effective solution for fine-tuning large language models (LLMs), but it often produces overconfident predictions in data-scarce few-shot settings. To address this issue, several classical statistical learning approaches have been repurposed for scalable uncertainty-aware LoRA fine-tuning. However, these approaches neglect how input characteristics affect the predictive uncertainty estimates. To address this limitation, we propose Contextual Low-Rank Adaptation (\textbf{C-LoRA}) as a novel uncertainty-aware and parameter efficient fine-tuning approach, by developing new lightweight LoRA modules contextualized to each input data sample to dynamically adapt uncertainty estimates. Incorporating data-driven contexts into the parameter posteriors, C-LoRA mitigates overfitting, achieves well-calibrated uncertainties, and yields robust predictions. Extensive experiments demonstrate that C-LoRA consistently outperforms the state-of-the-art uncertainty-aware LoRA methods in both uncertainty quantification and model generalization. Ablation studies further confirm the critical role of our contextual modules in capturing sample-specific uncertainties. C-LoRA sets a new standard for robust, uncertainty-aware LLM fine-tuning in few-shot regimes.

Unveiling Knowledge Utilization Mechanisms in LLM-based Retrieval-Augmented Generation

Considering the inherent limitations of parametric knowledge in large language models (LLMs), retrieval-augmented generation (RAG) is widely employed to expand their knowledge scope. Since RAG has shown promise in knowledge-intensive tasks like open-domain question answering, its broader application to complex tasks and intelligent assistants has further advanced its utility. Despite this progress, the underlying knowledge utilization mechanisms of LLM-based RAG remain underexplored. In this paper, we present a systematic investigation of the intrinsic mechanisms by which LLMs integrate internal (parametric) and external (retrieved) knowledge in RAG scenarios. Specially, we employ knowledge stream analysis at the macroscopic level, and investigate the function of individual modules at the microscopic level. Drawing on knowledge streaming analyses, we decompose the knowledge utilization process into four distinct stages within LLM layers: knowledge refinement, knowledge elicitation, knowledge expression, and knowledge contestation. We further demonstrate that the relevance of passages guides the streaming of knowledge through these stages. At the module level, we introduce a new method, knowledge activation probability entropy (KAPE) for neuron identification associated with either internal or external knowledge. By selectively deactivating these neurons, we achieve targeted shifts in the LLM's reliance on one knowledge source over the other. Moreover, we discern complementary roles for multi-head attention and multi-layer perceptron layers during knowledge formation. These insights offer a foundation for improving interpretability and reliability in retrieval-augmented LLMs, paving the way for more robust and transparent generative solutions in knowledge-intensive domains.

Aux-Think: Exploring Reasoning Strategies for Data-Efficient Vision-Language Navigation

Vision-Language Navigation (VLN) is a critical task for developing embodied agents that can follow natural language instructions to navigate in complex real-world environments. Recent advances in VLN by large pretrained models have significantly improved generalization and instruction grounding compared to traditional approaches. However, the role of reasoning strategies in navigation-an action-centric, long-horizon task-remains underexplored, despite Chain-of-Thought (CoT) reasoning's demonstrated success in static tasks like visual question answering. To address this gap, we conduct the first systematic evaluation of reasoning strategies for VLN, including No-Think (direct action prediction), Pre-Think (reason before action), and Post-Think (reason after action). Surprisingly, our findings reveal the Inference-time Reasoning Collapse issue, where inference-time reasoning degrades navigation accuracy, highlighting the challenges of integrating reasoning into VLN. Based on this insight, we propose Aux-Think, a framework that trains models to internalize structured reasoning patterns through CoT supervision, while inferring action directly without reasoning in online prediction. To support this framework, we release R2R-CoT-320k, the first Chain-of-Thought annotated dataset for VLN. Extensive experiments show that Aux-Think reduces training effort greatly and achieves the best performance under the same data scale.

DREAM: Disentangling Risks to Enhance Safety Alignment in Multimodal Large Language Models

Multimodal Large Language Models (MLLMs) pose unique safety challenges due to their integration of visual and textual data, thereby introducing new dimensions of potential attacks and complex risk combinations. In this paper, we begin with a detailed analysis aimed at disentangling risks through step-by-step reasoning within multimodal inputs. We find that systematic multimodal risk disentanglement substantially enhances the risk awareness of MLLMs. Via leveraging the strong discriminative abilities of multimodal risk disentanglement, we further introduce \textbf{DREAM} (\textit{\textbf{D}isentangling \textbf{R}isks to \textbf{E}nhance Safety \textbf{A}lignment in \textbf{M}LLMs}), a novel approach that enhances safety alignment in MLLMs through supervised fine-tuning and iterative Reinforcement Learning from AI Feedback (RLAIF). Experimental results show that DREAM significantly boosts safety during both inference and training phases without compromising performance on normal tasks (namely oversafety), achieving a 16.17\% improvement in the SIUO safe\&effective score compared to GPT-4V. The data and code are available at https://github.com/Kizna1ver/DREAM.

LongWriter-V: Enabling Ultra-Long and High-Fidelity Generation in Vision-Language Models

Existing Large Vision-Language Models (LVLMs) can process inputs with context lengths up to 128k visual and text tokens, yet they struggle to generate coherent outputs beyond 1,000 words. We find that the primary limitation is the absence of long output examples during supervised fine-tuning (SFT). To tackle this issue, we introduce LongWriter-V-22k, a SFT dataset comprising 22,158 examples, each with multiple input images, an instruction, and corresponding outputs ranging from 0 to 10,000 words. Moreover, to achieve long outputs that maintain high-fidelity to the input images, we employ Direct Preference Optimization (DPO) to the SFT model. Given the high cost of collecting human feedback for lengthy outputs (e.g., 3,000 words), we propose IterDPO, which breaks long outputs into segments and uses iterative corrections to form preference pairs with the original outputs. Additionally, we develop MMLongBench-Write, a benchmark featuring six tasks to evaluate the long-generation capabilities of VLMs. Our 7B parameter model, trained with LongWriter-V-22k and IterDPO, achieves impressive performance on this benchmark, outperforming larger proprietary models like GPT-4o. Code and data: https://github.com/THU-KEG/LongWriter-V

Retrieving Filter Spectra in CNN for Explainable Sleep Stage Classification

Despite significant advances in deep learning-based sleep stage classification, the clinical adoption of automatic classification models remains slow. One key challenge is the lack of explainability, as many models function as black boxes with millions of parameters. In response, recent work has increasingly focussed on enhancing model explainability. This study contributes to these efforts by globally explaining spectral processing of individual EEG channels. Specifically, we introduce a method to retrieve the filter spectrum of low-level convolutional feature extraction and compare it with the classification-relevant spectral information in the data. We evaluate our approach on the MSA-CNN model using the ISRUC-S3 and Sleep-EDF-20 datasets. Our findings show that spectral processing plays a significant role in the lower frequency bands. In addition, comparing the correlation between filter spectrum and data-based spectral information with univariate performance indicates that the model naturally prioritises the most informative channels in a multimodal setting. We specify how these insights can be leveraged to enhance model performance. The code for the filter spectrum retrieval and its analysis is available at https://github.com/sgoerttler/MSA-CNN.

MSA-CNN: A Lightweight Multi-Scale CNN with Attention for Sleep Stage Classification

Recent advancements in machine learning-based signal analysis, coupled with open data initiatives, have fuelled efforts in automatic sleep stage classification. Despite the proliferation of classification models, few have prioritised reducing model complexity, which is a crucial factor for practical applications. In this work, we introduce Multi-Scale and Attention Convolutional Neural Network (MSA-CNN), a lightweight architecture featuring as few as ~10,000 parameters. MSA-CNN leverages a novel multi-scale module employing complementary pooling to eliminate redundant filter parameters and dense convolutions. Model complexity is further reduced by separating temporal and spatial feature extraction and using cost-effective global spatial convolutions. This separation of tasks not only reduces model complexity but also mirrors the approach used by human experts in sleep stage scoring. We evaluated both small and large configurations of MSA-CNN against nine state-of-the-art baseline models across three public datasets, treating univariate and multivariate models separately. Our evaluation, based on repeated cross-validation and re-evaluation of all baseline models, demonstrated that the large MSA-CNN outperformed all baseline models on all three datasets in terms of accuracy and Cohen's kappa, despite its significantly reduced parameter count. Lastly, we explored various model variants and conducted an in-depth analysis of the key modules and techniques, providing deeper insights into the underlying mechanisms. The code for our models, baselines, and evaluation procedures is available at https://github.com/sgoerttler/MSA-CNN.

A Survey on Speech Large Language Models

Large Language Models (LLMs) exhibit strong contextual understanding and remarkable multi-task performance. Therefore, researchers have been seeking to integrate LLMs in the broad sense of Spoken Language Understanding (SLU) field. Different from the traditional method of cascading LLMs to process text generated by Automatic Speech Recognition(ASR), new efforts have focused on designing architectures centered around Audio Feature Extraction - Multimodal Information Fusion - LLM Inference(Speech LLMs). This approach enables richer audio feature extraction while simultaneously facilitating end-to-end fusion of audio and text modalities, thereby achieving deeper understanding and reasoning from audio data. This paper elucidates the development of Speech LLMs, offering an in-depth analysis of system architectures and training strategies. Through extensive research and a series of targeted experiments, the paper assesses Speech LLMs' advancements in Rich Audio Transcription and its potential for Cross-task Integration within the SLU field. Additionally, it indicates key challenges uncovered through experimentation, such as the Dormancy of LLMs under certain conditions. The paper further delves into the training strategies for Speech LLMs, proposing potential solutions based on these findings, and offering valuable insights and references for future research in this domain, as well as LLM applications in multimodal contexts.

Temporal Source Recovery for Time-Series Source-Free Unsupervised Domain Adaptation

Source-Free Unsupervised Domain Adaptation (SFUDA) has gained popularity for its ability to adapt pretrained models to target domains without accessing source domains, ensuring source data privacy. While SFUDA is well-developed in visual tasks, its application to Time-Series SFUDA (TS-SFUDA) remains limited due to the challenge of transferring crucial temporal dependencies across domains. Although a few researchers begin to explore this area, they rely on specific source domain designs, which are impractical as source data owners cannot be expected to follow particular pretraining protocols. To solve this, we propose Temporal Source Recovery (TemSR), a framework that transfers temporal dependencies for effective TS-SFUDA without requiring source-specific designs. TemSR features a recovery process that leverages masking, recovery, and optimization to generate a source-like distribution with recovered source temporal dependencies. To ensure effective recovery, we further design segment-based regularization to restore local dependencies and anchor-based recovery diversity maximization to enhance the diversity of the source-like distribution. The source-like distribution is then adapted to the target domain using traditional UDA techniques. Extensive experiments across multiple TS tasks demonstrate the effectiveness of TemSR, even surpassing existing TS-SFUDA method that requires source domain designs. Code is available in https://github.com/Frank-Wang-oss/TemSR.