Authors: Khaoula Chehbouni, Martine De Cock, Gilles Caporossi, Afaf Taik, Reihaneh Rabbany, Golnoosh Farnadi
Abstract: The increased screen time and isolation caused by the COVID-19 pandemic have led to a significant surge in cases of online grooming, which is the use of strategies by predators to lure children into sexual exploitation. Previous efforts to detect grooming in industry and academia have involved accessing and monitoring private conversations through centrally-trained models or sending private conversations to a global server. In this work, we implement a privacy-preserving pipeline for the early detection of sexual predators. We leverage federated learning and differential privacy in order to create safer online spaces for children while respecting their privacy. We investigate various privacy-preserving implementations and discuss their benefits and shortcomings. Our extensive evaluation using real-world data proves that privacy and utility can coexist with only a slight reduction in utility.
Authors: Juan Andres Medina Florez, Shaina Raza, Rashida Lynn, Zahra Shakeri, Brendan T. Smith, Elham Dolatabadi
Abstract: Understanding the prevalence, disparities, and symptom variations of Post COVID-19 Condition (PCC) for vulnerable populations is crucial to improving care and addressing intersecting inequities. This study aims to develop a comprehensive framework for integrating social determinants of health (SDOH) into PCC research by leveraging NLP techniques to analyze disparities and variations in SDOH representation within PCC case reports. Following construction of a PCC Case Report Corpus, comprising over 7,000 case reports from the LitCOVID repository, a subset of 709 reports were annotated with 26 core SDOH-related entity types using pre-trained named entity recognition (NER) models, human review, and data augmentation to improve quality, diversity and representation of entity types. An NLP pipeline integrating NER, natural language inference (NLI), trigram and frequency analyses was developed to extract and analyze these entities. Both encoder-only transformer models and RNN-based models were assessed for the NER objective. Fine-tuned encoder-only BERT models outperformed traditional RNN-based models in generalizability to distinct sentence structures and greater class sparsity. Exploratory analysis revealed variability in entity richness, with prevalent entities like condition, age, and access to care, and underrepresentation of sensitive categories like race and housing status. Trigram analysis highlighted frequent co-occurrences among entities, including age, gender, and condition. The NLI objective (entailment and contradiction analysis) showed attributes like "Experienced violence or abuse" and "Has medical insurance" had high entailment rates (82.4%-80.3%), while attributes such as "Is female-identifying," "Is married," and "Has a terminal condition" exhibited high contradiction rates (70.8%-98.5%).
Authors: Mikhail Krasitskii, Olga Kolesnikova, Liliana Chanona Hernandez, Grigori Sidorov, Alexander Gelbukh
Abstract: This study explores transformer-based models such as BERT, mBERT, and XLM-R for multi-lingual sentiment analysis across diverse linguistic structures. Key contributions include the identification of XLM-R superior adaptability in morphologically complex languages, achieving accuracy levels above 88%. The work highlights fine-tuning strategies and emphasizes their significance for improving sentiment classification in underrepresented languages.
Authors: Ningyu Xu, Qi Zhang, Chao Du, Qiang Luo, Xipeng Qiu, Xuanjing Huang, Menghan Zhang
Abstract: Recent advances in large language models (LLMs) provide a new opportunity to address the long-standing question of how concepts are represented and organized in the mind, which is central to unravelling the nature of human cognition. Here, we reframed the classic reverse dictionary task to simulate human concept inference in context and investigated the emergence of human-like conceptual representations within LLMs. We found that LLMs were able to infer concepts from definitional descriptions and construct representation spaces that converge towards a shared, context-independent structure. These representations effectively predicted human behavioural judgments and aligned well with neural activity patterns in the human brain, offering evidence for biological plausibility. These findings demonstrate that human-like conceptual representations and organization can naturally emerge from language prediction, even without real-world grounding. Our work supports the view that LLMs serve as valuable tools for understanding complex human cognition and paves the way for better alignment between artificial and human intelligence.
Authors: Haotian Luo, Li Shen, Haiying He, Yibo Wang, Shiwei Liu, Wei Li, Naiqiang Tan, Xiaochun Cao, Dacheng Tao
Abstract: Recently, long-thought reasoning LLMs, such as OpenAI's O1, adopt extended reasoning processes similar to how humans ponder over complex problems. This reasoning paradigm significantly enhances the model's problem-solving abilities and has achieved promising results. However, long-thought reasoning process leads to a substantial increase in inference time. A pressing challenge is reducing the inference overhead of long-thought LLMs while ensuring accuracy. In this paper, we experimentally demonstrate that long-thought reasoning models struggle to effectively allocate token budgets based on problem difficulty and reasoning redundancies. To address this, we propose Length-Harmonizing Fine-Tuning (O1-Pruner), aiming at minimizing reasoning overhead while maintaining accuracy. This effective fine-tuning method first estimates the LLM's baseline performance through pre-sampling and then uses RL-style fine-tuning to encourage the model to generate shorter reasoning processes under accuracy constraints. This allows the model to achieve efficient reasoning with lower redundancy while maintaining accuracy. Experiments on various mathematical reasoning benchmarks show that O1-Pruner not only significantly reduces inference overhead but also achieves higher accuracy, providing a novel and promising solution to this challenge. Our code is coming soon at https://github.com/StarDewXXX/O1-Pruner
Authors: Guodong Ma, Wenxuan Wang, Lifeng Zhou, Yuting Yang, Yuke Li, Binbin Du
Abstract: Recently, the Mixture of Expert (MoE) architecture, such as LR-MoE, is often used to alleviate the impact of language confusion on the multilingual ASR (MASR) task. However, it still faces language confusion issues, especially in mismatched domain scenarios. In this paper, we decouple language confusion in LR-MoE into confusion in self-attention and router. To alleviate the language confusion in self-attention, based on LR-MoE, we propose to apply attention-MoE architecture for MASR. In our new architecture, MoE is utilized not only on feed-forward network (FFN) but also on self-attention. In addition, to improve the robustness of the LID-based router on language confusion, we propose expert pruning and router augmentation methods. Combining the above, we get the boosted language-routing MoE (BLR-MoE) architecture. We verify the effectiveness of the proposed BLR-MoE in a 10,000-hour MASR dataset.
Authors: Lijun Li, Zhelun Shi, Xuhao Hu, Bowen Dong, Yiran Qin, Xihui Liu, Lu Sheng, Jing Shao
Abstract: Text-to-image (T2I) models have rapidly advanced, enabling the generation of high-quality images from text prompts across various domains. However, these models present notable safety concerns, including the risk of generating harmful, biased, or private content. Current research on assessing T2I safety remains in its early stages. While some efforts have been made to evaluate models on specific safety dimensions, many critical risks remain unexplored. To address this gap, we introduce T2ISafety, a safety benchmark that evaluates T2I models across three key domains: toxicity, fairness, and bias. We build a detailed hierarchy of 12 tasks and 44 categories based on these three domains, and meticulously collect 70K corresponding prompts. Based on this taxonomy and prompt set, we build a large-scale T2I dataset with 68K manually annotated images and train an evaluator capable of detecting critical risks that previous work has failed to identify, including risks that even ultra-large proprietary models like GPTs cannot correctly detect. We evaluate 12 prominent diffusion models on T2ISafety and reveal several concerns including persistent issues with racial fairness, a tendency to generate toxic content, and significant variation in privacy protection across the models, even with defense methods like concept erasing. Data and evaluator are released under https://github.com/adwardlee/t2i_safety.
Authors: Sunbowen Lee, Junting Zhou, Chang Ao, Kaige Li, Xinrun Du, Sirui He, Jiaheng Liu, Min Yang, Zhoufutu Wen, Shiwen Ni
Abstract: Model distillation is a technique for transferring knowledge from large language models (LLMs) to smaller ones, aiming to create resource-efficient yet high-performing models. However, excessive distillation can lead to homogenization, reducing diversity among models and impairing their ability to robustly handle complex or novel tasks. These limitations underscore the need to systematically quantify the distillation process and its impact. In this work, we propose a framework to evaluate and quantify model distillation. Our method addresses two key aspects: (1) Identifying identity cognition contradictions to assess discrepancies in how models perceive and represent identity-related information, and (2) Analyzing multi-granularity response similarities across models to measure the extent of homogenization. Experimental results demonstrate two key insights: (1) Well-known closed-source and open-source LLMs usually exhibit high distillation degrees, except for Claude, Doubao, and Gemini. (2) Base LLMs show higher distillation degrees compared to aligned LLMs. By offering a systematic approach to improve the transparency of LLM data distillation, we call for LLMs with more independent development and more transparent technical reports to improve LLMs' robustness and safety. The code and data are available under https://github.com/Aegis1863/LLMs-Distillation-Quantification.
URLs: https://github.com/Aegis1863/LLMs-Distillation-Quantification.
Authors: Naquee Rizwan, Nayandeep Deb, Sarthak Roy, Vishwajeet Singh Solanki, Kiran Garimella, Animesh Mukherjee
Abstract: Toxicity in digital media poses significant challenges, yet little attention has been given to its dynamics within the rapidly growing medium of podcasts. This paper addresses this gap by analyzing political podcast data to study the emergence and propagation of toxicity, focusing on conversation chains-structured reply patterns within podcast transcripts. Leveraging state-of-the-art transcription models and advanced conversational analysis techniques, we systematically examine toxic discourse in over 30 popular political podcasts in the United States. Our key contributions include: (1) creating a comprehensive dataset of transcribed and diarized political podcasts, identifying thousands of toxic instances using Google's Perspective API, (2) uncovering concerning trends where a majority of episodes contain at least one toxic instance, (3) introducing toxic conversation chains and analyzing their structural and linguistic properties, revealing characteristics such as longer durations, repetitive patterns, figurative language, and emotional cues tied to anger and annoyance, (4) identifying demand-related words like 'want', 'like', and 'know' as precursors to toxicity, and (5) developing predictive models to anticipate toxicity shifts based on annotated change points. Our findings provide critical insights into podcast toxicity and establish a foundation for future research on real-time monitoring and intervention mechanisms to foster healthier discourse in this influential medium.
Authors: Raj Sanjay Shah, Sashank Varma
Abstract: Many studies have evaluated the cognitive alignment of Pre-trained Language Models (PLMs), i.e., their correspondence to adult performance across a range of cognitive domains. Recently, the focus has expanded to the developmental alignment of these models: identifying phases during training where improvements in model performance track improvements in children's thinking over development. However, there are many challenges to the use of PLMs as cognitive science theories, including different architectures, different training data modalities and scales, and limited model interpretability. In this paper, we distill lessons learned from treating PLMs, not as engineering artifacts but as cognitive science and developmental science models. We review assumptions used by researchers to map measures of PLM performance to measures of human performance. We identify potential pitfalls of this approach to understanding human thinking, and we end by enumerating criteria for using PLMs as credible accounts of cognition and cognitive development.
Authors: Jan Christian Blaise Cruz, Alham Fikri Aji
Abstract: In this paper, we propose the use of simple knowledge distillation to produce smaller and more efficient single-language transformers from Massively Multilingual Transformers (MMTs) to alleviate tradeoffs associated with the use of such in low-resource settings. Using Tagalog as a case study, we show that these smaller single-language models perform on-par with strong baselines in a variety of benchmark tasks in a much more efficient manner. Furthermore, we investigate additional steps during the distillation process that improves the soft-supervision of the target language, and provide a number of analyses and ablations to show the efficacy of the proposed method.
Authors: Kai Yoshida, Masahiro Mizukami, Seiya Kawano, Canasai Kruengkrai, Hiroaki Sugiyama, Koichiro Yoshino
Abstract: To improve user engagement during conversations with dialogue systems, we must improve individual dialogue responses and dialogue impressions such as consistency, personality, and empathy throughout the entire dialogue. While such dialogue systems have been developing rapidly with the help of large language models (LLMs), reinforcement learning from AI feedback (RLAIF) has attracted attention to align LLM-based dialogue models for such dialogue impressions. In RLAIF, a reward model based on another LLM is used to create a training signal for an LLM-based dialogue model using zero-shot/few-shot prompting techniques. However, evaluating an entire dialogue only by prompting LLMs is challenging. In this study, the supervised fine-tuning (SFT) of LLMs prepared reward models corresponding to 12 metrics related to the impression of the entire dialogue for evaluating dialogue responses. We tuned our dialogue models using the reward model signals as feedback to improve the impression of the system. The results of automatic and human evaluations showed that tuning the dialogue model using our reward model corresponding to dialogue impression improved the evaluation of individual metrics and the naturalness of the dialogue response.
Authors: Chang Zong, Jian Wan, Lei Zhang
Abstract: Current LLM-based approaches improve question answering performance by leveraging the internal reasoning abilities of models or incorporating external knowledge. However, when humans address professional problems, it is essential to explicitly analyze the multifaceted relationships from multiple pieces and diverse sources of evidence to achieve better answers. In this study, we propose a novel generative question answering framework for the biomedical domain, named EvidenceMap, which explicitly learns and incorporates evidence analysis with small language models (SLMs). The framework describes an evidence map for each question and fully utilizes an SLM to derive the representation of the supportive evaluation, the logical correlation, and the summarization of the related evidence, which facilitates an analysis-augmented generation with another SLM in an autoregressive way. Extensive experiments have shown that introducing an evidence analysis learning process can significantly outperform larger models and popular LLM reasoning methods.
Authors: Chaochen Gao, Xing Wu, Zijia Lin, Debing Zhang, Songlin Hu
Abstract: Large language models (LLMs) with extended context windows have made significant strides yet remain a challenge due to the scarcity of long documents. Existing methods tend to synthesize long-context data but lack a clear mechanism to reinforce the long-range dependency modeling. To address this limitation, we propose NExtLong, a novel framework for synthesizing long-context data through Negative document Extension. NExtLong decomposes a document into multiple meta-chunks and extends the context by interleaving hard negative distractors retrieved from pretraining corpora. This approach compels the model to discriminate long-range dependent context from distracting content, enhancing its ability to model long-range dependencies. Extensive experiments demonstrate that NExtLong achieves significant performance improvements on the HELMET and RULER benchmarks compared to existing long-context synthesis approaches and leading models, which are trained on non-synthetic long documents. These findings highlight NExtLong's ability to reduce reliance on non-synthetic long documents, making it an effective framework for developing advanced long-context LLMs.
Authors: Seyed Mahed Mousavi, Simone Alghisi, Giuseppe Riccardi
Abstract: LLMs' sources of knowledge are data snapshots containing factual information about entities collected at different timestamps and from different media types (e.g. wikis, social media, etc.). Such unstructured knowledge is subject to change due to updates through time from past to present. Equally important are the inconsistencies and inaccuracies occurring in different information sources. Consequently, the model's knowledge about an entity may be perturbed while training over the sequence of snapshots or at inference time, resulting in inconsistent and inaccurate model performance. In this work, we study the appropriateness of Large Language Models (LLMs) as repositories of factual knowledge. We consider twenty-four state-of-the-art LLMs that are either closed-, partially (weights), or fully (weight and training data) open-source. We evaluate their reliability in responding to time-sensitive factual questions in terms of accuracy and consistency when prompts are perturbed. We further evaluate the effectiveness of state-of-the-art methods to improve LLMs' accuracy and consistency. We then propose "ENtity-Aware Fine-tuning" (ENAF), a soft neurosymbolic approach aimed at providing a structured representation of entities during fine-tuning to improve the model's performance.
Authors: Duc Hau Nguyen, Cyrielle Mallart, Guillaume Gravier, Pascale S\'ebillot
Abstract: Attention mechanism is contributing to the majority of recent advances in machine learning for natural language processing. Additionally, it results in an attention map that shows the proportional influence of each input in its decision. Empirical studies postulate that attention maps can be provided as an explanation for model output. However, it is still questionable to ask whether this explanation helps regular people to understand and accept the model output (the plausibility of the explanation). Recent studies show that attention weights in the RNN encoders are hardly plausible because they spread on input tokens. We thus propose 3 additional constraints to the learning objective function to improve the plausibility of the attention map: regularization to increase the attention weight sparsity, semi-supervision to supervise the map by a heuristic and supervision by human annotation. Results show that all techniques can improve the attention map plausibility at some level. We also observe that specific instructions for human annotation might have a negative effect on classification performance. Beyond the attention map, the result of experiments on text classification tasks also shows that no matter how the constraint brings the gain, the contextualization layer plays a crucial role in finding the right space for finding plausible tokens.
Authors: Simone Filice, Guy Horowitz, David Carmel, Zohar Karnin, Liane Lewin-Eytan, Yoelle Maarek
Abstract: Evaluating Retrieval-Augmented Generation (RAG) systems, especially in domain-specific contexts, requires benchmarks that address the distinctive requirements of the applicative scenario. Since real data can be hard to obtain, a common strategy is to use LLM-based methods to generate synthetic data. Existing solutions are general purpose: given a document, they generate a question to build a Q&A pair. However, although the generated questions can be individually good, they are typically not diverse enough to reasonably cover the different ways real end-users can interact with the RAG system. We introduce here DataMorgana, a tool for generating highly customizable and diverse synthetic Q&A benchmarks tailored to RAG applications. DataMorgana enables detailed configurations of user and question categories and provides control over their distribution within the benchmark. It uses a lightweight two-stage process, ensuring efficiency and fast iterations, while generating benchmarks that reflect the expected traffic. We conduct a thorough line of experiments, showing quantitatively and qualitatively that DataMorgana surpasses existing tools and approaches in producing lexically, syntactically, and semantically diverse question sets across domain-specific and general-knowledge corpora. DataMorgana will be made available to selected teams in the research community, as first beta testers, in the context of the upcoming SIGIR'2025 LiveRAG challenge to be announced in early February 2025.
Authors: Felix Buend\'ia, Joaqu\'in Gayoso-Cabada, Jos\'e-Luis Sierra
Abstract: In this paper, we describe an approach to transforming the huge amount of medical knowledge available in existing online medical collections into standardized learning packages ready to be integrated into the most popular e-learning platforms. The core of our approach is a tool called Clavy, which makes it possible to retrieve pieces of content in medical collections, to transform this content into meaningful learning units, and to export it in the form of standardized learning packages. In addition to describing the approach, we demonstrate its feasibility by applying it to the generation of IMS content packages from MedPix, a popular online database of medical cases in the domain of radiology.
Authors: John Pavlopoulos, Juli Bakagianni, Kanella Pouli, Maria Gavriilidou
Abstract: Natural Language Processing (NLP) for lesser-resourced languages faces persistent challenges, including limited datasets, inherited biases from high-resource languages, and the need for domain-specific solutions. This study addresses these gaps for Modern Greek through three key contributions. First, we evaluate the performance of open-source (Llama-70b) and closed-source (GPT-4o mini) large language models (LLMs) on seven core NLP tasks with dataset availability, revealing task-specific strengths, weaknesses, and parity in their performance. Second, we expand the scope of Greek NLP by reframing Authorship Attribution as a tool to assess potential data usage by LLMs in pre-training, with high 0-shot accuracy suggesting ethical implications for data provenance. Third, we showcase a legal NLP case study, where a Summarize, Translate, and Embed (STE) methodology outperforms the traditional TF-IDF approach for clustering \emph{long} legal texts. Together, these contributions provide a roadmap to advance NLP in lesser-resourced languages, bridging gaps in model evaluation, task innovation, and real-world impact.
Authors: Viktor Moskvoretskii, Maria Lysyuk, Mikhail Salnikov, Nikolay Ivanov, Sergey Pletenev, Daria Galimzianova, Nikita Krayko, Vasily Konovalov, Irina Nikishina, Alexander Panchenko
Abstract: Retrieval Augmented Generation (RAG) improves correctness of Question Answering (QA) and addresses hallucinations in Large Language Models (LLMs), yet greatly increase computational costs. Besides, RAG is not always needed as may introduce irrelevant information. Recent adaptive retrieval methods integrate LLMs' intrinsic knowledge with external information appealing to LLM self-knowledge, but they often neglect efficiency evaluations and comparisons with uncertainty estimation techniques. We bridge this gap by conducting a comprehensive analysis of 35 adaptive retrieval methods, including 8 recent approaches and 27 uncertainty estimation techniques, across 6 datasets using 10 metrics for QA performance, self-knowledge, and efficiency. Our findings show that uncertainty estimation techniques often outperform complex pipelines in terms of efficiency and self-knowledge, while maintaining comparable QA performance.
Authors: Chen Chen, Xinlong Hao, Weiwen Liu, Xu Huang, Xingshan Zeng, Shuai Yu, Dexun Li, Shuai Wang, Weinan Gan, Yuefeng Huang, Xinzhi Wang, Defu Lian, Baoqun Yin, Yasheng Wang, Wu Liu
Abstract: Large language models (LLMs) have demonstrated significant potential in decision-making and reasoning, especially when combined with various tools to effectively solve complex problems. However, existing evaluation systems for assessing LLM function calling capabilities have several limitations: (1) limited evaluation scenarios, lacking assessments in real multi-turn dialogue contexts; (2) narrow evaluation dimensions, lacking detailed assessments for fine-grained function calls; (3) relying on LLMs or real API executions for result evaluation, which introduces significant overhead. To address these issues, we propose a comprehensive evaluation system named ACEBench. This system is meticulously designed to encompass a wide spectrum of function calling scenarios. Moreover, it categorizes these scenarios into three primary types according to the evaluation methodology: Normal, Special, and Agent. Normal evaluates function calls in basic scenarios; Special evaluates function calls in scenarios with vague or incomplete instructions; Agent introduces multi-agent interactions to simulate function calling evaluation in real-world multi-turn interactions. We conducted extensive experiments on ACEBench, analyzing various LLMs in-depth and performing a more granular analysis of error causes across different data types.
Authors: Jingyuan Chen, Tao Wu, Wei Ji, Fei Wu
Abstract: Large language models (LLMs) have emerged as powerful tools in natural language processing (NLP), showing a promising future of artificial generated intelligence (AGI). Despite their notable performance in the general domain, LLMs have remained suboptimal in the field of education, owing to the unique challenges presented by this domain, such as the need for more specialized knowledge, the requirement for personalized learning experiences, and the necessity for concise explanations of complex concepts. To address these issues, this paper presents a novel LLM for education named WisdomBot, which combines the power of LLMs with educational theories, enabling their seamless integration into educational contexts. To be specific, we harness self-instructed knowledge concepts and instructions under the guidance of Bloom's Taxonomy as training data. To further enhance the accuracy and professionalism of model's response on factual questions, we introduce two key enhancements during inference, i.e., local knowledge base retrieval augmentation and search engine retrieval augmentation during inference. We substantiate the effectiveness of our approach by applying it to several Chinese LLMs, thereby showcasing that the fine-tuned models can generate more reliable and professional responses.
Authors: Yafu Li, Xuyang Hu, Xiaoye Qu, Linjie Li, Yu Cheng
Abstract: Large language models (LLMs) demonstrate impressive performance but lack the flexibility to adapt to human preferences quickly without retraining. In this work, we introduce Test-time Preference Optimization (TPO), a framework that aligns LLM outputs with human preferences during inference, removing the need to update model parameters. Rather than relying on purely numerical rewards, TPO translates reward signals into textual critiques and uses them as textual rewards to iteratively refine its response. Evaluations on benchmarks covering instruction following, preference alignment, safety, and mathematics reveal that TPO progressively improves alignment with human preferences. Notably, after only a few TPO steps, the initially unaligned Llama-3.1-70B-SFT model can surpass the aligned counterpart, Llama-3.1-70B-Instruct. Furthermore, TPO scales efficiently with both the search width and depth during inference. Through case studies, we illustrate how TPO exploits the innate capacity of LLM to interpret and act upon reward signals. Our findings establish TPO as a practical, lightweight alternative for test-time preference optimization, achieving alignment on the fly. Our code is publicly available at https://github.com/yafuly/TPO.
Authors: Offa Kingsleigh, Alfred Abercrombie, David Woolstencroft, Beorhtric Meadowcroft, Marcus Irvin
Abstract: Contextual Partitioning introduces an innovative approach to enhancing the architectural design of large-scale computational models through the dynamic segmentation of parameters into context-aware regions. This methodology emphasizes the importance of task-specific specialization, achieved through adaptive parameter allocation mechanisms that align with the linguistic features of input data. Experimental evaluations demonstrated substantial improvements in accuracy, perplexity, and contextual coherence across a variety of linguistic tasks, highlighting the adaptability and scalability of the proposed framework. By reducing redundancy and enhancing computational efficiency, Contextual Partitioning not only streamlines model operations but also expands the scope of applications for advanced language processing systems. The approach operates autonomously, requiring no external fine-tuning, thereby addressing a significant limitation in conventional parameter optimization techniques. Empirical results demonstrate the effectiveness of gradient-driven segmentation, enabling models to dynamically recalibrate and specialize in response to task-specific demands. Furthermore, resource utilization metrics reveal notable reductions in memory usage and training times, confirming the efficiency of the approach. Observations from qualitative analyses illustrate improved contextual coherence and logical flow in generated outputs, reinforcing the practical value of this technique. The findings collectively demonstrate the potential for Contextual Partitioning to redefine the scalability and adaptability of computational language architectures in diverse and complex domains.
Authors: Zhenran Xu, Longyue Wang, Jifang Wang, Zhouyi Li, Senbao Shi, Xue Yang, Yiyu Wang, Baotian Hu, Jun Yu, Min Zhang
Abstract: Virtual film production requires intricate decision-making processes, including scriptwriting, virtual cinematography, and precise actor positioning and actions. Motivated by recent advances in automated decision-making with language agent-based societies, this paper introduces FilmAgent, a novel LLM-based multi-agent collaborative framework for end-to-end film automation in our constructed 3D virtual spaces. FilmAgent simulates various crew roles, including directors, screenwriters, actors, and cinematographers, and covers key stages of a film production workflow: (1) idea development transforms brainstormed ideas into structured story outlines; (2) scriptwriting elaborates on dialogue and character actions for each scene; (3) cinematography determines the camera setups for each shot. A team of agents collaborates through iterative feedback and revisions, thereby verifying intermediate scripts and reducing hallucinations. We evaluate the generated videos on 15 ideas and 4 key aspects. Human evaluation shows that FilmAgent outperforms all baselines across all aspects and scores 3.98 out of 5 on average, showing the feasibility of multi-agent collaboration in filmmaking. Further analysis reveals that FilmAgent, despite using the less advanced GPT-4o model, surpasses the single-agent o1, showing the advantage of a well-coordinated multi-agent system. Lastly, we discuss the complementary strengths and weaknesses of OpenAI's text-to-video model Sora and our FilmAgent in filmmaking.
Authors: Joaqu\'i Gayoso-Cabada, Mar\'ia Goicoechea-de-Jorge, Mercedes G\'omez-Albarr\'an, Amelia Sanz-Cabrerizo, Antonio Sarasa-Cabezuelo, Jos\'e-Luis Sierra
Abstract: Information and communications technology and technology-enhanced learning have unquestionably transformed traditional teaching-learning processes and are positioned as key factors to promote quality education, one of the basic sustainable development goals of the 2030 agenda. Document annotation, which was traditionally carried out with pencil and paper and currently benefits from digital document annotation tools, is a representative example of this transformation. Using document annotation tools, students can enrich the documents with annotations that highlight the most relevant aspects of these documents. As the conceptual complexity of the learning domain increases, the annotation of the documents may require comprehensive domain knowledge and an expert analysis capability that students usually lack. Consequently, a proliferation of irrelevant, incorrect, and/or poorly decontextualized annotations may appear, while other relevant aspects are completely ignored by the students. The main hypothesis proposed by this paper is that the use of a guiding annotation ontology in the annotation activities is a keystone aspect to alleviate these shortcomings. Consequently, comprehension is improved, exhaustive content analysis is promoted, and meta-reflective thinking is developed. To test this hypothesis, we describe our own annotation tool, \@note, which fully implements this ontology-enhanced annotation paradigm, and we provide experimental evidence about how \@note can improve academic performance via a pilot study concerning critical literary annotation.
Authors: DeepSeek-AI, Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, Aixin Liu, Bing Xue, Bingxuan Wang, Bochao Wu, Bei Feng, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, Damai Dai, Deli Chen, Dongjie Ji, Erhang Li, Fangyun Lin, Fucong Dai, Fuli Luo, Guangbo Hao, Guanting Chen, Guowei Li, H. Zhang, Han Bao, Hanwei Xu, Haocheng Wang, Honghui Ding, Huajian Xin, Huazuo Gao, Hui Qu, Hui Li, Jianzhong Guo, Jiashi Li, Jiawei Wang, Jingchang Chen, Jingyang Yuan, Junjie Qiu, Junlong Li, J. L. Cai, Jiaqi Ni, Jian Liang, Jin Chen, Kai Dong, Kai Hu, Kaige Gao, Kang Guan, Kexin Huang, Kuai Yu, Lean Wang, Lecong Zhang, Liang Zhao, Litong Wang, Liyue Zhang, Lei Xu, Leyi Xia, Mingchuan Zhang, Minghua Zhang, Minghui Tang, Meng Li, Miaojun Wang, Mingming Li, Ning Tian, Panpan Huang, Peng Zhang, Qiancheng Wang, Qinyu Chen, Qiushi Du, Ruiqi Ge, Ruisong Zhang, Ruizhe Pan, Runji Wang, R. J. Chen, R. L. Jin, Ruyi Chen, Shanghao Lu, Shangyan Zhou, Shanhuang Chen, Shengfeng Ye, Shiyu Wang, Shuiping Yu, Shunfeng Zhou, Shuting Pan, S. S. Li, Shuang Zhou, Shaoqing Wu, Shengfeng Ye, Tao Yun, Tian Pei, Tianyu Sun, T. Wang, Wangding Zeng, Wanjia Zhao, Wen Liu, Wenfeng Liang, Wenjun Gao, Wenqin Yu, Wentao Zhang, W. L. Xiao, Wei An, Xiaodong Liu, Xiaohan Wang, Xiaokang Chen, Xiaotao Nie, Xin Cheng, Xin Liu, Xin Xie, Xingchao Liu, Xinyu Yang, Xinyuan Li, Xuecheng Su, Xuheng Lin, X. Q. Li, Xiangyue Jin, Xiaojin Shen, Xiaosha Chen, Xiaowen Sun, Xiaoxiang Wang, Xinnan Song, Xinyi Zhou, Xianzu Wang, Xinxia Shan, Y. K. Li, Y. Q. Wang, Y. X. Wei, Yang Zhang, Yanhong Xu, Yao Li, Yao Zhao, Yaofeng Sun, Yaohui Wang, Yi Yu, Yichao Zhang, Yifan Shi, Yiliang Xiong, Ying He, Yishi Piao, Yisong Wang, Yixuan Tan, Yiyang Ma, Yiyuan Liu, Yongqiang Guo, Yuan Ou, Yuduan Wang, Yue Gong, Yuheng Zou, Yujia He, Yunfan Xiong, Yuxiang Luo, Yuxiang You, Yuxuan Liu, Yuyang Zhou, Y. X. Zhu, Yanhong Xu, Yanping Huang, Yaohui Li, Yi Zheng, Yuchen Zhu, Yunxian Ma, Ying Tang, Yukun Zha, Yuting Yan, Z. Z. Ren, Zehui Ren, Zhangli Sha, Zhe Fu, Zhean Xu, Zhenda Xie, Zhengyan Zhang, Zhewen Hao, Zhicheng Ma, Zhigang Yan, Zhiyu Wu, Zihui Gu, Zijia Zhu, Zijun Liu, Zilin Li, Ziwei Xie, Ziyang Song, Zizheng Pan, Zhen Huang, Zhipeng Xu, Zhongyu Zhang, Zhen Zhang
Abstract: We introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1. DeepSeek-R1-Zero, a model trained via large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step, demonstrates remarkable reasoning capabilities. Through RL, DeepSeek-R1-Zero naturally emerges with numerous powerful and intriguing reasoning behaviors. However, it encounters challenges such as poor readability, and language mixing. To address these issues and further enhance reasoning performance, we introduce DeepSeek-R1, which incorporates multi-stage training and cold-start data before RL. DeepSeek-R1 achieves performance comparable to OpenAI-o1-1217 on reasoning tasks. To support the research community, we open-source DeepSeek-R1-Zero, DeepSeek-R1, and six dense models (1.5B, 7B, 8B, 14B, 32B, 70B) distilled from DeepSeek-R1 based on Qwen and Llama.
Authors: Krzysztof Bartnicki, Stanis{\l}aw Dro\.zd\.z, Jaros{\l}aw Kwapie\'n, Tomasz Stanisz
Abstract: The complexity characteristics of texts written in natural languages are significantly related to the rules of punctuation. In particular, the distances between punctuation marks measured by the number of words quite universally follow the family of Weibull distributions known from survival analyses. However, the values of two parameters marking specific forms of these distributions distinguish specific languages. This is such a strong constraint that the punctuation distributions of texts translated from the original language into another adopt quantitative characteristics of the target language. All these changes take place within Weibull distributions such that the corresponding hazard functions are always increasing. Recent previous research shows that James Joyce's famous "Finnegans Wake" is subject to such extreme distribution from the Weibull family that the corresponding hazard function is clearly decreasing. At the same time, the distances of sentence ending punctuation marks, determining the variability of sentence length, have an almost perfect multifractal organization, so far to such an extent found nowhere else in the literature. In the present contribution based on several available translations (Dutch, French, German, Polish, Russian) of "Finnegans Wake", it is shown that the punctuation characteristics of this work remain largely translation invariant, contrary to the common cases. These observations may constitute further evidence that "Finnegans Wake" is a translinguistic work in this respect as well, in line with Joyce's original intention.
Authors: Jakub Dec, Micha{\l} Dolina, Stanis{\l}aw Dro\.zd\.z, Jaros{\l}aw Kwapie\'n, Tomasz Stanisz
Abstract: Punctuation is the main factor introducing correlations in natural language written texts and it crucially impacts their overall effectiveness, expressiveness, and readability. Punctuation marks at the end of sentences are of particular importance as their distribution can determine various complexity features of written natural language. Here, the sentence length variability (SLV) time series representing "Hopscotch" by Julio Cortazar are subjected to quantitative analysis with an attempt to identify their distribution type, long-memory effects, and potential multiscale patterns. The analyzed novel is an important and innovative piece of literature whose essential property is freedom of movement between its building blocks given to a reader by the author. The statistical consequences of this freedom are closely investigated in both the original, Spanish version of the novel, and its translations into English and Polish. Clear evidence of rich multifractality in the SLV dynamics, with a left-sided asymmetry, however, is observed in all three language versions as well as in the versions with differently ordered chapters.
Authors: Weizhi Fei, Xueyan Niu, Guoqing Xie, Yingqing Liu, Bo Bai, Wei Han
Abstract: Although applications involving long-context inputs are crucial for the effective utilization of large language models (LLMs), they also result in increased computational costs and reduced performance. To address this challenge, we propose an efficient, training-free prompt compression method that retains key information within compressed prompts. We identify specific attention heads in transformer-based LLMs, which we designate as evaluator heads, that are capable of selecting tokens in long inputs that are most significant for inference. Building on this discovery, we develop EHPC, an Evaluator Head-based Prompt Compression method, which enables LLMs to rapidly "skim through" input prompts by leveraging only the first few layers with evaluator heads during the pre-filling stage, subsequently passing only the important tokens to the model for inference. EHPC achieves state-of-the-art results across two mainstream benchmarks: prompt compression and long-context inference acceleration. Consequently, it effectively reduces the complexity and costs associated with commercial API calls. We further demonstrate that EHPC attains competitive results compared to key-value cache-based acceleration methods, thereby highlighting its potential to enhance the efficiency of LLMs for long-context tasks.
Authors: Chongren Sun, Yuran Li, Di Wu, Benoit Boulet
Abstract: Large Language Models (LLMs) are highly capable but require significant computational resources for both training and inference. Within the LLM family, smaller models (those with fewer than 10 billion parameters) also perform well across various tasks. However, these smaller models share similar limitations to their larger counterparts, including the tendency to hallucinate. Despite the existence of many benchmarks to evaluate hallucination in LLMs, few have specifically focused on small LLMs (SLLMs). Additionally, SLLMs show widely varying performance across different benchmarks. In this paper, we introduce OnionEval, a multi-layer structured framework with a specific metric called the context-influence score (CI), designed to effectively assess the fact-conflicting hallucination tendencies of small LLMs across different contextual levels. Our experimental results reveal a key feature of SLLMs: they excel in factual analysis but face challenges with context reasoning. Further investigation shows that a simple Chain-of-Thought strategy can significantly reduce these limitations, improving the practical usefulness of SLLMs in real-world applications.
Authors: Moreno La Quatra, Valerio Mario Salerno, Yu Tsao, Sabato Marco Siniscalchi
Abstract: In this paper, we present an encoder-decoder model leveraging Flan-T5 for post-Automatic Speech Recognition (ASR) Generative Speech Error Correction (GenSEC), and we refer to it as FlanEC. We explore its application within the GenSEC framework to enhance ASR outputs by mapping n-best hypotheses into a single output sentence. By utilizing n-best lists from ASR models, we aim to improve the linguistic correctness, accuracy, and grammaticality of final ASR transcriptions. Specifically, we investigate whether scaling the training data and incorporating diverse datasets can lead to significant improvements in post-ASR error correction. We evaluate FlanEC using the HyPoradise dataset, providing a comprehensive analysis of the model's effectiveness in this domain. Furthermore, we assess the proposed approach under different settings to evaluate model scalability and efficiency, offering valuable insights into the potential of instruction-tuned encoder-decoder models for this task.
Authors: Florian Kankowski, Torgrim Solstad, Sina Zarriess, Oliver Bott
Abstract: In this paper, we compare data generated with mono- and multilingual LLMs spanning a range of model sizes with data provided by human participants in an experimental setting investigating well-established discourse biases. Beyond the comparison as such, we aim to develop a benchmark to assess the capabilities of LLMs with discourse biases as a robust proxy for more general discourse understanding capabilities. More specifically, we investigated Implicit Causality verbs, for which psycholinguistic research has found participants to display biases with regard to three phenomena:\ the establishment of (i) coreference relations (Experiment 1), (ii) coherence relations (Experiment 2), and (iii) the use of particular referring expressions (Experiments 3 and 4). With regard to coreference biases we found only the largest monolingual LLM (German Bloom 6.4B) to display more human-like biases. For coherence relation, no LLM displayed the explanation bias usually found for humans. For referring expressions, all LLMs displayed a preference for referring to subject arguments with simpler forms than to objects. However, no bias effect on referring expression was found, as opposed to recent studies investigating human biases.
Authors: Yantao Liu, Zijun Yao, Rui Min, Yixin Cao, Lei Hou, Juanzi Li
Abstract: Best-of-N (BoN) sampling, a common strategy for test-time scaling of Large Language Models (LLMs), relies on reward models to select the best candidate solution from multiple generations. However, traditional reward models often assign arbitrary and inconsistent scores, limiting their effectiveness. To address this, we propose a Pairwise Reward Model (Pairwise RM) combined with a knockout tournament for BoN sampling. Instead of assigning absolute scores, given one math problem, Pairwise RM evaluates two candidate solutions' correctness simultaneously. This approach eliminates the need for arbitrary scoring and enables cross-validation of solutions through parallel comparison. In the knockout tournament, Pairwise RM conducts pairwise comparisons between candidate solutions and eliminates the incorrect ones iteratively. We construct \ourdataset, a large-scale dataset of 443K pairwise comparisons derived from NumiaMath and annotated using \texttt{gemini-1.5-flash}, and train the Pairwise RM via supervised fine-tuning. Experiments on MATH-500 and the Olympiad Bench demonstrate significant improvements over traditional discriminative reward models. And a 40\% to 60\% relative improvement is achieved on the top 50\% challenging problems.
Authors: Bohao Yang, Yingji Zhang, Dong Liu, Andr\'e Freitas, Chenghua Lin
Abstract: Recent large language models (LLMs) have advanced table understanding capabilities but rely on converting tables into text sequences. While multimodal large language models (MLLMs) enable direct visual processing, they face limitations in handling scientific tables due to fixed input image resolutions and insufficient numerical reasoning capabilities. We present a comprehensive framework for multimodal scientific table understanding and reasoning with dynamic input image resolutions. Our framework consists of three key components: (1) MMSci-Pre, a domain-specific table structure learning dataset of 52K scientific table structure recognition samples, (2) MMSci-Ins, an instruction tuning dataset with 12K samples across three table-based tasks, and (3) MMSci-Eval, a benchmark with 3,114 testing samples specifically designed to evaluate numerical reasoning capabilities. Extensive experiments demonstrate that our domain-specific approach with 52K scientific table images achieves superior performance compared to 150K general-domain tables, highlighting the importance of data quality over quantity. Our proposed table-based MLLMs with dynamic input resolutions show significant improvements in both general table understanding and numerical reasoning capabilities, with strong generalisation to held-out datasets. Our code and data are publicly available at https://github.com/Bernard-Yang/MMSci_Table.
Authors: Ang Lv, Ruobing Xie, Yining Qian, Songhao Wu, Xingwu Sun, Zhanhui Kang, Di Wang, Rui Yan
Abstract: Mixture-of-Experts (MoE) models mostly use a router to assign tokens to specific expert modules, activating only partial parameters and often outperforming dense models. We argue that the separation between the router's decision-making and the experts' execution is a critical yet overlooked issue, leading to suboptimal expert selection and ineffective learning. To address this, we propose Autonomy-of-Experts (AoE), a novel MoE paradigm in which experts autonomously select themselves to process inputs. AoE is based on the insight that an expert is aware of its own capacity to effectively process a token, an awareness reflected in the scale of its internal activations. In AoE, routers are removed; instead, experts pre-compute internal activations for inputs and are ranked based on their activation norms. Only the top-ranking experts proceed with the forward pass, while the others abort. The overhead of pre-computing activations is reduced through a low-rank weight factorization. This self-evaluating-then-partner-comparing approach ensures improved expert selection and effective learning. We pre-train language models having 700M up to 4B parameters, demonstrating that AoE outperforms traditional MoE models with comparable efficiency.
Authors: Melissa Kazemi Rad, Huy Nghiem, Andy Luo, Sahil Wadhwa, Mohammad Sorower, Stephen Rawls
Abstract: Large Language Models (LLMs) have demonstrated powerful capabilities that render them valuable in different applications, including conversational AI products. It is paramount to ensure the security and reliability of these products by mitigating their vulnerabilities towards malicious user interactions, which can lead to the exposure of great risks and reputational repercussions. In this work, we present a comprehensive study on the efficacy of fine-tuning and aligning Chain-of-Thought (CoT) responses of different LLMs that serve as input moderation guardrails. We systematically explore various tuning methods by leveraging a small set of training data to adapt these models as proxy defense mechanisms to detect malicious inputs and provide a reasoning for their verdicts, thereby preventing the exploitation of conversational agents. We rigorously evaluate the efficacy and robustness of different tuning strategies to generalize across diverse adversarial and malicious query types. Our experimental results outline the potential of alignment processes tailored to a varied range of harmful input queries, even with constrained data resources. These techniques significantly enhance the safety of conversational AI systems and provide a feasible framework for deploying more secure and trustworthy AI-driven interactions.
Authors: M. Manzour, A. Ballardini, R. Izquierdo, M. \'A. Sotelo
Abstract: Lane-changing maneuvers, particularly those executed abruptly or in risky situations, are a significant cause of road traffic accidents. However, current research mainly focuses on predicting safe lane changes. Furthermore, existing accident datasets are often based on images only and lack comprehensive sensory data. In this work, we focus on predicting risky lane changes using the CRASH dataset (our own collected dataset specifically for risky lane changes), and safe lane changes (using the HighD dataset). Then, we leverage KG and Bayesian inference to predict these maneuvers using linguistic contextual information, enhancing the model's interpretability and transparency. The model achieved a 91.5% f1-score with anticipation time extending to four seconds for risky lane changes, and a 90.0% f1-score for predicting safe lane changes with the same anticipation time. We validate our model by integrating it into a vehicle within the CARLA simulator in scenarios that involve risky lane changes. The model managed to anticipate sudden lane changes, thus providing automated vehicles with further time to plan and execute appropriate safe reactions. Finally, to enhance the explainability of our model, we utilize RAG to provide clear and natural language explanations for the given prediction.
Authors: Shijie Han, Changhai Zhou, Yiqing Shen, Tianning Sun, Yuhua Zhou, Xiaoxia Wang, Zhixiao Yang, Jingshu Zhang, Hongguang Li
Abstract: Current financial Large Language Models (LLMs) struggle with two critical limitations: a lack of depth in stock analysis, which impedes their ability to generate professional-grade insights, and the absence of objective evaluation metrics to assess the quality of stock analysis reports. To address these challenges, this paper introduces FinSphere, a conversational stock analysis agent, along with three major contributions: (1) Stocksis, a dataset curated by industry experts to enhance LLMs' stock analysis capabilities, (2) AnalyScore, a systematic evaluation framework for assessing stock analysis quality, and (3) FinSphere, an AI agent that can generate high-quality stock analysis reports in response to user queries. Experiments demonstrate that FinSphere achieves superior performance compared to both general and domain-specific LLMs, as well as existing agent-based systems, even when they are enhanced with real-time data access and few-shot guidance. The integrated framework, which combines real-time data feeds, quantitative tools, and an instruction-tuned LLM, yields substantial improvements in both analytical quality and practical applicability for real-world stock analysis.
Authors: Kush R. Varshney, Zahra Ashktorab, Djallel Bouneffouf, Matthew Riemer, Justin D. Weisz
Abstract: Much of the research focus on AI alignment seeks to align large language models and other foundation models to the context-less and generic values of helpfulness, harmlessness, and honesty. Frontier model providers also strive to align their models with these values. In this paper, we motivate why we need to move beyond such a limited conception and propose three dimensions for doing so. The first scope of alignment is competence: knowledge, skills, or behaviors the model must possess to be useful for its intended purpose. The second scope of alignment is transience: either semantic or episodic depending on the context of use. The third scope of alignment is audience: either mass, public, small-group, or dyadic. At the end of the paper, we use the proposed framework to position some technologies and workflows that go beyond prevailing notions of alignment.
Authors: Yifan Liu, Yaokun Liu, Zelin Li, Ruichen Yao, Yang Zhang, Dong Wang
Abstract: The proliferation of fake news on social media platforms disproportionately impacts vulnerable populations, eroding trust, exacerbating inequality, and amplifying harmful narratives. Detecting fake news in multimodal contexts -- where deceptive content combines text and images -- is particularly challenging due to the nuanced interplay between modalities. Existing multimodal fake news detection methods often emphasize cross-modal consistency but ignore the complex interactions between text and visual elements, which may complement, contradict, or independently influence the predicted veracity of a post. To address these challenges, we present Modality Interactive Mixture-of-Experts for Fake News Detection (MIMoE-FND), a novel hierarchical Mixture-of-Experts framework designed to enhance multimodal fake news detection by explicitly modeling modality interactions through an interaction gating mechanism. Our approach models modality interactions by evaluating two key aspects of modality interactions: unimodal prediction agreement and semantic alignment. The hierarchical structure of MIMoE-FND allows for distinct learning pathways tailored to different fusion scenarios, adapting to the unique characteristics of each modality interaction. By tailoring fusion strategies to diverse modality interaction scenarios, MIMoE-FND provides a more robust and nuanced approach to multimodal fake news detection. We evaluate our approach on three real-world benchmarks spanning two languages, demonstrating its superior performance compared to state-of-the-art methods. By enhancing the accuracy and interpretability of fake news detection, MIMoE-FND offers a promising tool to mitigate the spread of misinformation, with the potential to better safeguard vulnerable communities against its harmful effects.
Authors: Dongsheng Zhu, Weixian Shi, Zhengliang Shi, Zhaochun Ren, Shuaiqiang Wang, Lingyong Yan, Dawei Yin
Abstract: Although current Large Language Models (LLMs) exhibit impressive capabilities, performing complex real-world tasks still requires tool learning. Mainstream methods, such as CoT/ReAct, rely on step-by-step tool invocation to interact with external environments, but they are limited in perceptual scope and lack adequate task-planning capability. To address these limitations, other studies introduce the first Search-based Decision Tree (DFSDT), which still suffers from the high computational cost. In this paper, we introduce a novel parallel tool invocation paradigm, DTA-Llama (Divide-Then-Aggregate Llama). First, we transform traditional tree-based tool search paths into Directed Acyclic Graph (DAG) structure, generating a high-quality parallel tool invocation dataset. The DTA-Llama is then trained on the dataset to learn to iteratively divide the current task into several parallel tool invocation sub-tasks and aggregate the invocation results to decide the next actions. Furthermore, we introduce an efficient inference framework inspired by the Process/Threads mechanism when applying the DTA-Llama to practical tasks. Experimental results show that our approach substantially enhances task performance while reducing token consumption and inference time. Llama2-7B, using our method, is comparable to the official parallel function calling method of GPT-3.5. The relevant code, dataset, and model weights are available at https://corn0205.github.io/
Authors: Tabinda Aman, Mohammad Nadeem, Shahab Saquib Sohail, Mohammad Anas, Erik Cambria
Abstract: Animal stereotypes are deeply embedded in human culture and language. They often shape our perceptions and expectations of various species. Our study investigates how animal stereotypes manifest in vision-language models during the task of image generation. Through targeted prompts, we explore whether DALL-E perpetuates stereotypical representations of animals, such as "owls as wise," "foxes as unfaithful," etc. Our findings reveal significant stereotyped instances where the model consistently generates images aligned with cultural biases. The current work is the first of its kind to examine animal stereotyping in vision-language models systematically and to highlight a critical yet underexplored dimension of bias in AI-generated visual content.
Authors: Tian Jin, Ahmed Imtiaz Humayun, Utku Evci, Suvinay Subramanian, Amir Yazdanbakhsh, Dan Alistarh, Gintare Karolina Dziugaite
Abstract: Pruning eliminates unnecessary parameters in neural networks; it offers a promising solution to the growing computational demands of large language models (LLMs). While many focus on post-training pruning, sparse pre-training--which combines pruning and pre-training into a single phase--provides a simpler alternative. In this work, we present the first systematic exploration of optimal sparse pre-training configurations for LLMs through an examination of 80 unique pruning schedules across different sparsity levels and training durations. We find that initiating pruning at 25% of total training compute and concluding at 75% achieves near-optimal final evaluation loss. These findings provide valuable insights for efficient and effective sparse pre-training of LLMs. Furthermore, we propose a new scaling law that modifies the Chinchilla scaling law to use the average parameter count over pre-training. Through empirical and theoretical validation, we demonstrate that this modified scaling law accurately models evaluation loss for both sparsely and densely pre-trained LLMs, unifying scaling laws across pre-training paradigms. Our findings indicate that while sparse pre-training achieves the same final model quality as dense pre-training for equivalent compute budgets, it provides substantial benefits through reduced model size, enabling significant potential computational savings during inference.
Authors: Vanya Cohen, Geraud Nangue Tasse, Nakul Gopalan, Steven James, Matthew Gombolay, Ray Mooney, Benjamin Rosman
Abstract: Combining reinforcement learning with language grounding is challenging as the agent needs to explore the environment while simultaneously learning multiple language-conditioned tasks. To address this, we introduce a novel method: the compositionally-enabled reinforcement learning language agent (CERLLA). Our method reduces the sample complexity of tasks specified with language by leveraging compositional policy representations and a semantic parser trained using reinforcement learning and in-context learning. We evaluate our approach in an environment requiring function approximation and demonstrate compositional generalization to novel tasks. Our method significantly outperforms the previous best non-compositional baseline in terms of sample complexity on 162 tasks designed to test compositional generalization. Our model attains a higher success rate and learns in fewer steps than the non-compositional baseline. It reaches a success rate equal to an oracle policy's upper-bound performance of 92%. With the same number of environment steps, the baseline only reaches a success rate of 80%.
Authors: Rock Yuren Pang, Hope Schroeder, Kynnedy Simone Smith, Solon Barocas, Ziang Xiao, Emily Tseng, Danielle Bragg
Abstract: Large language models (LLMs) have been positioned to revolutionize HCI, by reshaping not only the interfaces, design patterns, and sociotechnical systems that we study, but also the research practices we use. To-date, however, there has been little understanding of LLMs' uptake in HCI. We address this gap via a systematic literature review of 153 CHI papers from 2020-24 that engage with LLMs. We taxonomize: (1) domains where LLMs are applied; (2) roles of LLMs in HCI projects; (3) contribution types; and (4) acknowledged limitations and risks. We find LLM work in 10 diverse domains, primarily via empirical and artifact contributions. Authors use LLMs in five distinct roles, including as research tools or simulated users. Still, authors often raise validity and reproducibility concerns, and overwhelmingly study closed models. We outline opportunities to improve HCI research with and on LLMs, and provide guiding questions for researchers to consider the validity and appropriateness of LLM-related work.
Authors: Enes Arda, Aylin Yener
Abstract: This paper introduces an information-theoretic framework for text summarization. We define the summarizer rate-distortion function and show that it provides a fundamental lower bound on summarizer performance. We describe an iterative procedure, similar to Blahut-Arimoto algorithm, for computing this function. To handle real-world text datasets, we also propose a practical method that can calculate the summarizer rate-distortion function with limited data. Finally, we empirically confirm our theoretical results by comparing the summarizer rate-distortion function with the performances of different summarizers used in practice.
Authors: Hessa Abdulrahman Alawwad, Areej Alhothali, Usman Naseem, Ali Alkhathlan, Amani Jamal
Abstract: Textbook question answering (TQA) is a challenging task in artificial intelligence due to the complex nature of context needed to answer complex questions. Although previous research has improved the task, there are still some limitations in textual TQA, including weak reasoning and inability to capture contextual information in the lengthy context. We propose a framework (PLRTQA) that incorporates the retrieval augmented generation (RAG) technique to handle the out-of-domain scenario where concepts are spread across different lessons, and utilize transfer learning to handle the long context and enhance reasoning abilities. Our architecture outperforms the baseline, achieving an accuracy improvement of 4. 12% in the validation set and 9. 84% in the test set for textual multiple-choice questions. While this paper focuses on solving challenges in the textual TQA, It provides a foundation for future work in multimodal TQA where the visual components are integrated to address more complex educational scenarios. Code: https://github.com/hessaAlawwad/PLR-TQA
Authors: Xin Xu, Shizhe Diao, Can Yang, Yang Wang
Abstract: Chain-of-Thought (CoT) prompting has marked a significant advancement in enhancing the reasoning capabilities of large language models (LLMs). Previous studies have developed various extensions of CoT, which focus primarily on enhancing end-task performance. In addition, there has been research on assessing the quality of reasoning chains in CoT. This raises an intriguing question: Is it possible to predict the accuracy of LLM outputs by scrutinizing the reasoning chains they generate? To answer this research question, we introduce a benchmark, R2PE, designed specifically to explore the relationship between reasoning chains and performance in various reasoning tasks spanning five different domains. This benchmark aims to measure the falsehood of the final output of LLMs based on the reasoning steps. To make full use of information in multiple reasoning chains, we propose the process discernibility score (PDS) framework that beats the answer-checking baseline by a large margin. Concretely, this resulted in an average of $5.1\%$ increase in the F1 score and $2.97\%$ improvement in AUC-PR across all 45 subsets within R2PE. We further demonstrate our PDS's efficacy in advancing open-domain QA accuracy.
Authors: Qi Gou, Cam-Tu Nguyen
Abstract: Large Language Models (LLMs) have become increasingly popular due to their ability to process and generate natural language. However, as they are trained on massive datasets of text, LLMs can inherit harmful biases and produce outputs that are not aligned with human values. This paper studies two main approaches to LLM alignment: Reinforcement Learning with Human Feedback (RLHF) and contrastive learning-based methods like Direct Preference Optimization (DPO). By analyzing the stability and robustness of RLHF and DPO, we propose MPO (Mixed Preference Optimization), a novel method that mitigates the weaknesses of both approaches. Specifically, we propose a two-stage training procedure: first train DPO on an easy dataset, and then perform RLHF on a difficult set with DPO model being the reference model. Here, the easy and difficult sets are constructed by a well-trained reward model that splits response pairs into those with large gaps of reward (easy), and those with small gaps (difficult). The first stage allows us to obtain a relatively optimal policy (LLM) model quickly, whereas the second stage refines LLM with online RLHF, thus mitigating the distribution shift issue associated with DPO. Experiments are conducted on two public alignment datasets, namely HH-RLHF and TLDR, demonstrating the effectiveness of MPO, both in terms of GPT4 and human evaluation.
Authors: Sophie Ostmeier, Justin Xu, Zhihong Chen, Maya Varma, Louis Blankemeier, Christian Bluethgen, Arne Edward Michalson, Michael Moseley, Curtis Langlotz, Akshay S Chaudhari, Jean-Benoit Delbrouck
Abstract: Evaluating radiology reports is a challenging problem as factual correctness is extremely important due to the need for accurate medical communication about medical images. Existing automatic evaluation metrics either suffer from failing to consider factual correctness (e.g., BLEU and ROUGE) or are limited in their interpretability (e.g., F1CheXpert and F1RadGraph). In this paper, we introduce GREEN (Generative Radiology Report Evaluation and Error Notation), a radiology report generation metric that leverages the natural language understanding of language models to identify and explain clinically significant errors in candidate reports, both quantitatively and qualitatively. Compared to current metrics, GREEN offers: 1) a score aligned with expert preferences, 2) human interpretable explanations of clinically significant errors, enabling feedback loops with end-users, and 3) a lightweight open-source method that reaches the performance of commercial counterparts. We validate our GREEN metric by comparing it to GPT-4, as well as to error counts of 6 experts and preferences of 2 experts. Our method demonstrates not only higher correlation with expert error counts, but simultaneously higher alignment with expert preferences when compared to previous approaches.
Authors: Armand Nicolicioiu, Eugenia Iofinova, Eldar Kurtic, Mahdi Nikdan, Andrei Panferov, Ilia Markov, Nir Shavit, Dan Alistarh
Abstract: The availability of powerful open-source large language models (LLMs) opens exciting use cases, such as automated personal assistants that adapt to the user's unique data and demands. Two key requirements for such assistants are personalization - in the sense that the assistant should reflect the user's own writing style - and privacy - users may prefer to always store their personal data locally, on their own computing device. In this application paper, we present a new design and evaluation for such an automated assistant, for the specific use case of email generation, which we call Panza. Specifically, Panza can be trained and deployed locally on commodity hardware, and is personalized to the user's writing style. Panza's personalization features are based on a combination of fine-tuning using a variant of the Reverse Instructions technique together with Retrieval-Augmented Generation (RAG). We demonstrate that this combination allows us to fine-tune an LLM to better reflect a user's writing style using limited data, while executing on extremely limited resources, e.g. on a free Google Colab instance. Our key methodological contribution is what we believe to be the first detailed study of evaluation metrics for this personalized writing task, and of how different choices of system components - e.g. the use of RAG and of different fine-tuning approaches - impact the system's performance. We are releasing the full Panza code as well as a new "David" personalized email dataset licensed for research use, both available on https://github.com/IST-DASLab/PanzaMail.
Authors: Jonathan Bourne
Abstract: The digitisation of historical print media archives is crucial for increasing accessibility to contemporary records. However, the process of Optical Character Recognition (OCR) used to convert physical records to digital text is prone to errors, particularly in the case of newspapers and periodicals due to their complex layouts. This paper introduces Context Leveraging OCR Correction (CLOCR-C), which utilises the infilling and context-adaptive abilities of transformer-based language models (LMs) to improve OCR quality. The study aims to determine if LMs can perform post-OCR correction, improve downstream NLP tasks, and the value of providing the socio-cultural context as part of the correction process. Experiments were conducted using seven LMs on three datasets: the 19th Century Serials Edition (NCSE) and two datasets from the Overproof collection. The results demonstrate that some LMs can significantly reduce error rates, with the top-performing model achieving over a 60\% reduction in character error rate on the NCSE dataset. The OCR improvements extend to downstream tasks, such as Named Entity Recognition, with increased Cosine Named Entity Similarity. Furthermore, the study shows that providing socio-cultural context in the prompts improves performance, while misleading prompts lower performance. In addition to the findings, this study releases a dataset of 91 transcribed articles from the NCSE, containing a total of 40 thousand words, to support further research in this area. The findings suggest that CLOCR-C is a promising approach for enhancing the quality of existing digital archives by leveraging the socio-cultural information embedded in the LMs and the text requiring correction.
Authors: Xindi Tong, Yujin Zhu, Shijian Fan, Liang Xu
Abstract: Long text summarization, gradually being essential for efficiently processing large volumes of information, stays challenging for Large Language Models (LLMs) such as GPT and LLaMA families because of the insufficient open-sourced training datasets and the high requirement of contextual details dealing. To address the issue, we design a novel zero-shot transfer learning framework, abbreviated as T3, to iteratively training a baseline LLM on an assistant task for the target task, where the former should own richer data resources and share structural or semantic similarity with the latter. In practice, T3 is approached to deal with the long text summarization task by utilizing question answering as the assistant task, and further validated its effectiveness on the BBC summary, NarraSum, FairytaleQA, and NLQuAD datasets, with up to nearly 14% improvement in ROUGE, 35% improvement in BLEU, and 16% improvement in Factscore compared to three baseline LLMs, demonstrating its potential for more assistant-target task combinations.
Authors: Noel Elias, Homa Esfahanizadeh, Kaan Kale, Sriram Vishwanath, Muriel Medard
Abstract: Large language models have drastically changed the prospects of AI by introducing technologies for more complex natural language processing. However, current methodologies to train such LLMs require extensive resources including but not limited to large amounts of data, expensive machinery, and lengthy training. To solve this problem, this paper proposes a new tokenization method inspired by universal Lempel-Ziv-Welch data compression that compresses repetitive phrases into multi-word tokens. With MultiTok as a new tokenizing tool, we show that language models are able to be trained notably more efficiently while offering a similar accuracy on more succinct and compressed training data. In fact, our results demonstrate that MultiTok achieves a comparable performance to the BERT and GPT-2 standards as both a stand-alone tokenizer and an add-on to existing tokenizers while also providing close to 2.5x faster training with more than 30% less training data.
Authors: Rina Carines Cabral, Soyeon Caren Han, Areej Alhassan, Riza Batista-Navarro, Goran Nenadic, Josiah Poon
Abstract: Discontinuous Named Entity Recognition (DNER) presents a challenging problem where entities may be scattered across multiple non-adjacent tokens, making traditional sequence labelling approaches inadequate. Existing methods predominantly rely on custom tagging schemes to handle these discontinuous entities, resulting in models tightly coupled to specific tagging strategies and lacking generalisability across diverse datasets. To address these challenges, we propose TriG-NER, a novel Triplet-Grid Framework that introduces a generalisable approach to learning robust token-level representations for discontinuous entity extraction. Our framework applies triplet loss at the token level, where similarity is defined by word pairs existing within the same entity, effectively pulling together similar and pushing apart dissimilar ones. This approach enhances entity boundary detection and reduces the dependency on specific tagging schemes by focusing on word-pair relationships within a flexible grid structure. We evaluate TriG-NER on three benchmark DNER datasets and demonstrate significant improvements over existing grid-based architectures. These results underscore our framework's effectiveness in capturing complex entity structures and its adaptability to various tagging schemes, setting a new benchmark for discontinuous entity extraction.
Authors: Shashi Kumar, Iuliia Thorbecke, Sergio Burdisso, Esa\'u Villatoro-Tello, Manjunath K E, Kadri Hacio\u{g}lu, Pradeep Rangappa, Petr Motlicek, Aravind Ganapathiraju, Andreas Stolcke
Abstract: Recent research has demonstrated that training a linear connector between speech foundation encoders and large language models (LLMs) enables this architecture to achieve strong ASR capabilities. Despite the impressive results, it remains unclear whether these simple approaches are robust enough across different scenarios and speech conditions, such as domain shifts and speech perturbations. In this paper, we address these questions by conducting various ablation experiments using a recent and widely adopted approach called SLAM-ASR. We present novel empirical findings that offer insights on how to effectively utilize the SLAM-ASR architecture across a wide range of settings. Our main findings indicate that SLAM-ASR exhibits poor performance in cross-domain evaluation settings. Additionally, speech perturbations on in-domain data, such as changes in speech rate or additive noise, can significantly degrade performance. Our findings offer critical insights for fine-tuning and configuring robust LLM-based ASR models, tailored to different data characteristics and computational resources.
Authors: Nan Xu, Xuezhe Ma
Abstract: While large language models (LLMs) have demonstrated increasing power, they have also called upon studies on their hallucinated outputs that deviate from factually correct statements. In this paper, we focus on one important scenario of false premises, where LLMs are distracted by misaligned claims although the model possesses the required factual knowledge to answer original questions accurately. Inspired by the observation that entropy of the false-premise prompt is closely related to its likelihood to elicit hallucination generation, we propose a new prompting algorithm, named DecoPrompt, to mitigate hallucination. DecoPrompt leverages LLMs to "decode" the false-premise prompts without really eliciting hallucination output from LLMs. We perform experiments on two datasets, demonstrating that DecoPrompt can reduce hallucinations effectively on outputs from different LLMs. Moreover, DecoPrompt exhibits cross-model transferability, which facilitates its applications to scenarios such as LLMs of large sizes or unavailable model logits.
Authors: Alan Wake, Bei Chen, C. X. Lv, Chao Li, Chengen Huang, Chenglin Cai, Chujie Zheng, Daniel Cooper, Fan Zhou, Feng Hu, Ge Zhang, Guoyin Wang, Heng Ji, Howard Qiu, Jiangcheng Zhu, Jun Tian, Katherine Su, Lihuan Zhang, Liying Li, Ming Song, Mou Li, Peng Liu, Qicheng Hu, Shawn Wang, Shijun Zhou, Shiming Yang, Shiyong Li, Tianhang Zhu, Wen Xie, Wenhao Huang, Xiang He, Xiaobo Chen, Xiaohui Hu, Xiaoyi Ren, Xinyao Niu, Yanpeng Li, Yongke Zhao, Yongzhen Luo, Yuchi Xu, Yuxuan Sha, Zhaodong Yan, Zhiyuan Liu, Zirui Zhang, Zonghong Dai
Abstract: This technical report presents Yi-Lightning, our latest flagship large language model (LLM). It achieves exceptional performance, ranking 6th overall on Chatbot Arena, with particularly strong results (2nd to 4th place) in specialized categories including Chinese, Math, Coding, and Hard Prompts. Yi-Lightning leverages an enhanced Mixture-of-Experts (MoE) architecture, featuring advanced expert segmentation and routing mechanisms coupled with optimized KV-caching techniques. Our development process encompasses comprehensive pre-training, supervised fine-tuning (SFT), and reinforcement learning from human feedback (RLHF), where we devise deliberate strategies for multi-stage training, synthetic data construction, and reward modeling. Furthermore, we implement RAISE (Responsible AI Safety Engine), a four-component framework to address safety issues across pre-training, post-training, and serving phases. Empowered by our scalable super-computing infrastructure, all these innovations substantially reduce training, deployment and inference costs while maintaining high-performance standards. With further evaluations on public academic benchmarks, Yi-Lightning demonstrates competitive performance against top-tier LLMs, while we observe a notable disparity between traditional, static benchmark results and real-world, dynamic human preferences. This observation prompts a critical reassessment of conventional benchmarks' utility in guiding the development of more intelligent and powerful AI systems for practical applications. Yi-Lightning is now available through our developer platform at https://platform.lingyiwanwu.com.
Authors: Javier de la Rosa, Vladislav Mikhailov, Lemei Zhang, Freddy Wetjen, David Samuel, Peng Liu, Rolv-Arild Braaten, Petter M{\ae}hlum, Magnus Breder Birkenes, Andrey Kutuzov, Tita Enstad, Hans Christian Farseth{\aa}s, Svein Arne Brygfjeld, Jon Atle Gulla, Stephan Oepen, Erik Velldal, Wilfred {\O}stgulen, Liljia {\O}vrelid, Aslak Sira Myhre
Abstract: The use of copyrighted materials in training language models raises critical legal and ethical questions. This paper presents a framework for and the results of empirically assessing the impact of publisher-controlled copyrighted corpora on the performance of generative large language models (LLMs) for Norwegian. When evaluated on a diverse set of tasks, we found that adding both books and newspapers to the data mixture of LLMs tend to improve their performance, while the addition of fiction works seems to be detrimental. Our experiments could inform the creation of a compensation scheme for authors whose works contribute to AI development.
Authors: Ruixi Lin, Yang You
Abstract: One of the potential failures of large language models (LLMs) is their imbalanced class performances in text classification tasks. With in-context learning (ICL), LLMs yields good accuracy for some classes but low accuracy for others. This imbalance is particularly problematic when misclassifications lead to user dissatisfaction or safety risks. While the root causes may lie in the data, addressing them from the source through training is neither easy nor cost-effective. To delve deeper, the imbalance stems from certain classes consistently receiving disproportionately high ICL probabilities, while others receive lower probabilities, resulting in under-prediction and lower accuracy in the latter. Crucially, probability ranges vary in their impact on the imbalance, enabling precise corrections by range. Therefore, this work introduces an inference-time debiasing method, FuRud (Fuzzy Rule Optimization-based Debiasing), to tackle this issue. FuRud addresses core interpretability challenges by determining why certain classes require corrections and tailoring adjustments for each sample and class probability. Tailored corrections use fuzzy sets with triangular membership functions, because they can transform per-sample class probabilities based on probability ranges. Each class selects one from 19 triangular membership functions, solving a nonlinear integer programming selection problem with simulated annealing, to minimize class accuracy bias (COBias) and maximize overall accuracy without updating LLM parameters. Notably, across seven benchmark datasets, FuRud reduces COBias by more than half (56%), while achieving a relative increase of 21% in overall accuracy, outperforming state-of-the-art debiasing methods.
Authors: Ao Sun
Abstract: In this work, we introduce CHAIR (Classifier of Hallucination As ImproveR), a supervised framework for detecting hallucinations by analyzing internal logits from each layer of every token. Our method extracts a compact set of features such as maximum, minimum, mean, standard deviation, and slope-from the token logits across all layers, enabling effective hallucination detection without overfitting. Experiments on TruthfulQA and MMLU datasets demonstrate that CHAIR significantly improves detection accuracy, particularly in zero-shot scenarios, showcasing its robustness and generalizability. Beyond hallucination detection, CHAIR highlights the potential of using internal representations for designing advanced decoding strategies. By leveraging patterns in logits, we suggest that more sophisticated models and adaptive decoding methods could further reduce hallucinations and enhance text completion quality. CHAIR not only offers a practical solution for detecting hallucinations but also lays the groundwork for exploring richer representations in LLMs to improve their factuality and coherence.
Authors: Pablo Miralles-Gonz\'alez, Javier Huertas-Tato, Alejandro Mart\'in, David Camacho
Abstract: The rapid advancement in large language models (LLMs) has significantly enhanced their ability to generate coherent and contextually relevant text, raising concerns about the misuse of AI-generated content and making it critical to detect it. However, the task remains challenging, particularly in unseen domains or with unfamiliar LLMs. Leveraging LLM next-token distribution outputs offers a theoretically appealing approach for detection, as they encapsulate insights from the models' extensive pre-training on diverse corpora. Despite its promise, zero-shot methods that attempt to operationalize these outputs have met with limited success. We hypothesize that one of the problems is that they use the mean to aggregate next-token distribution metrics across tokens, when some tokens are naturally easier or harder to predict and should be weighted differently. Based on this idea, we propose the Perplexity Attention Weighted Network (PAWN), which uses the last hidden states of the LLM and positions to weight the sum of a series of features based on metrics from the next-token distribution across the sequence length. Although not zero-shot, our method allows us to cache the last hidden states and next-token distribution metrics on disk, greatly reducing the training resource requirements. PAWN shows competitive and even better performance in-distribution than the strongest baselines (fine-tuned LMs) with a fraction of their trainable parameters. Our model also generalizes better to unseen domains and source models, with smaller variability in the decision boundary across distribution shifts. It is also more robust to adversarial attacks, and if the backbone has multilingual capabilities, it presents decent generalization to languages not seen during supervised training, with LLaMA3-1B reaching a mean macro-averaged F1 score of 81.46% in cross-validation with nine languages.
Authors: Xin Huang, Tarun Kumar Vangani, Minh Duc Pham, Xunlong Zou, Bin Wang, Zhengyuan Liu, Ai Ti Aw
Abstract: Multilingual large language models (MLLMs) have shown impressive capabilities across a variety of languages. However, efficacy can differ greatly between different language families, especially for those with limited linguistic resources. This report presents MERaLiON-TextLLM, a series of open-source language models specifically tailored to improve understanding and generation in Chinese, Indonesian, Malay, and Singlish. The initial released model is built on Llama-3-8B-Base and refined through a meticulously crafted process of continued pre-training and weight merging. Our approach achieves performance improvements across benchmarks in these languages, exceeding the capabilities of the official Llama-3 models. We provide the model checkpoints as a resource to support further research and development in cross-lingual language understanding.
Authors: Zengyi Gao, Yukun Cao, Hairu Wang, Ao Ke, Yuan Feng, Xike Xie, S Kevin Zhou
Abstract: To mitigate the hallucination and knowledge deficiency in large language models (LLMs), Knowledge Graph (KG)-based Retrieval-Augmented Generation (RAG) has shown promising potential by utilizing KGs as external resource to enhance LLMs reasoning. However, existing KG-RAG approaches struggle with a trade-off between flexibility and retrieval quality. Modular methods prioritize flexibility by avoiding the use of KG-fine-tuned models during retrieval, leading to fixed retrieval strategies and suboptimal retrieval quality. Conversely, coupled methods embed KG information within models to improve retrieval quality, but at the expense of flexibility. In this paper, we propose a novel flexible modular KG-RAG framework, termed FRAG, which synergizes the advantages of both approaches. FRAG estimates the hop range of reasoning paths based solely on the query and classify it as either simple or complex. To match the complexity of the query, tailored pipelines are applied to ensure efficient and accurate reasoning path retrieval, thus fostering the final reasoning process. By using the query text instead of the KG to infer the structural information of reasoning paths and employing adaptable retrieval strategies, FRAG improves retrieval quality while maintaining flexibility. Moreover, FRAG does not require extra LLMs fine-tuning or calls, significantly boosting efficiency and conserving resources. Extensive experiments show that FRAG achieves state-of-the-art performance with high efficiency and low resource consumption.
Authors: Aarush Sinha, Viraj Virk, Dipshikha Chakraborty, P. S. Sreeja
Abstract: Language Models [LMs] are now playing an increasingly large role in information generation and synthesis; the representation of scientific knowledge in these systems needs to be highly accurate. A prime challenge is hallucination; that is, generating apparently plausible but actually false information, including invented citations and nonexistent research papers. This kind of inaccuracy is dangerous in all the domains that require high levels of factual correctness, such as academia and education. This work presents a pipeline for evaluating the frequency with which language models hallucinate in generating responses in the scientific literature. We propose ArxEval, an evaluation pipeline with two tasks using ArXiv as a repository: Jumbled Titles and Mixed Titles. Our evaluation includes fifteen widely used language models and provides comparative insights into their reliability in handling scientific literature.
Authors: Jingyuan Yang, Rongjun Li, Weixuan Wang, Ziyu Zhou, Zhiyong Feng, Wei Peng
Abstract: Large Language Models (LLMs) often generate inconsistent responses when prompted with semantically equivalent paraphrased inputs. Recently, activation steering, a technique that modulates LLMs' behaviours by adjusting their latent representations during inference time, has been explored to improve the semantic consistency of LLMs. However, these methods typically operate at the model component level, such as layer hidden states or attention head outputs. They face a challenge due to the ``polysemanticity issue'', where the model components of LLMs typically encode multiple entangled features, making precise steering difficult. To address this challenge, we drill down to feature-level representations and propose LF-Steering, a novel activation steering approach to precisely identify latent feature representations responsible for semantic inconsistency. More specifically, our method maps the hidden states of the relevant transformer layer into a sparsely activated, high-dimensional feature space based on a sparse autoencoder (SAE), ensuring model steering based on decoupled feature representations with minimal interference. Comprehensive experiments on NLU and NLG datasets demonstrate the effectiveness of our method in enhancing semantic consistency, resulting in significant performance gains for various NLU and NLG tasks.
Authors: Yuxing Lu, Sin Yee Goi, Xukai Zhao, Jinzhuo Wang
Abstract: Biomedical knowledge graphs (BKGs) have emerged as powerful tools for organizing and leveraging the vast and complex data found across the biomedical field. Yet, current reviews of BKGs often limit their scope to specific domains or methods, overlooking the broader landscape and the rapid technological progress reshaping it. In this survey, we address this gap by offering a systematic review of BKGs from three core perspectives: domains, tasks, and applications. We begin by examining how BKGs are constructed from diverse data sources, including molecular interactions, pharmacological datasets, and clinical records. Next, we discuss the essential tasks enabled by BKGs, focusing on knowledge management, retrieval, reasoning, and interpretation. Finally, we highlight real-world applications in precision medicine, drug discovery, and scientific research, illustrating the translational impact of BKGs across multiple sectors. By synthesizing these perspectives into a unified framework, this survey not only clarifies the current state of BKG research but also establishes a foundation for future exploration, enabling both innovative methodological advances and practical implementations.
Authors: Karishma Thakrar, Katrina Lawrence, Kyle Howard
Abstract: Stylistic text generation plays a vital role in enhancing communication by reflecting the nuances of individual expression. This paper presents a novel approach for generating text in a specific speaker's style across different languages. We show that by leveraging only 100 lines of text, an individuals unique style can be captured as a high-dimensional embedding, which can be used for both text generation and stylistic translation. This methodology breaks down the language barrier by transferring the style of a speaker between languages. The paper is structured into three main phases: augmenting the speaker's data with stylistically consistent external sources, separating style from content using machine learning and deep learning techniques, and generating an abstract style profile by mean pooling the learned embeddings. The proposed approach is shown to be topic-agnostic, with test accuracy and F1 scores of 74.9% and 0.75, respectively. The results demonstrate the potential of the style profile for multilingual communication, paving the way for further applications in personalized content generation and cross-linguistic stylistic transfer.
Authors: Keer Lu, Zheng Liang, Da Pan, Shusen Zhang, Xin Wu, Weipeng Chen, Zenan Zhou, Guosheng Dong, Bin Cui, Wentao Zhang
Abstract: In recent years, Large Language Models (LLMs) have exhibited remarkable capabilities in clinical scenarios. However, despite their potential, existing works face challenges when applying LLMs to medical settings. Strategies relying on training with medical datasets are highly cost-intensive and may suffer from outdated training data. Leveraging external knowledge bases is a suitable alternative, yet it faces obstacles such as limited retrieval precision and poor effectiveness in answer extraction. These issues collectively prevent LLMs from demonstrating the expected level of proficiency in mastering medical expertise. To address these challenges, we introduce Med-R^2, a novel LLM physician framework that adheres to the Evidence-Based Medicine (EBM) process, efficiently integrating retrieval mechanisms as well as the selection and reasoning processes of evidence, thereby enhancing the problem-solving capabilities of LLMs in healthcare scenarios and fostering a trustworthy LLM physician. Our comprehensive experiments indicate that Med-R^2 achieves a 14.87\% improvement over vanilla RAG methods and even a 3.59\% enhancement compared to fine-tuning strategies, without incurring additional training costs.
Authors: Rabiul Awal, Saba Ahmadi, Le Zhang, Aishwarya Agrawal
Abstract: Fine-grained understanding of objects, attributes, and relationships between objects is crucial for visual-language models (VLMs). Existing benchmarks primarily focus on evaluating VLMs' capability to distinguish between two very similar captions given an image. In this paper, we introduce a new, challenging benchmark termed Visual Minimal-Change Understanding (VisMin), which requires models to predict the correct image-caption match given two images and two captions. The image pair and caption pair contain minimal changes, i.e., only one aspect changes at a time from among the following: object, attribute, count, and spatial relation. These changes test the models' understanding of objects, attributes (such as color, material, shape), counts, and spatial relationships between objects. We built an automatic framework using large language models and diffusion models, followed by a rigorous 4-step verification process by human annotators. Empirical experiments reveal that current VLMs exhibit notable deficiencies in understanding spatial relationships and counting abilities. We also generate a large-scale training dataset to finetune CLIP and Idefics2, showing significant improvements in fine-grained understanding across benchmarks and in CLIP's general image-text alignment. We release all resources, including the benchmark, training data, and finetuned model checkpoints, at https://vismin.net/.
URLs: https://vismin.net/.
Authors: Rainer Muehlhoff, Marte Henningsen
Abstract: This study examines the AI-powered grading tool "AI Grading Assistant" by the German company Fobizz, designed to support teachers in evaluating and providing feedback on student assignments. Against the societal backdrop of an overburdened education system and rising expectations for artificial intelligence as a solution to these challenges, the investigation evaluates the tool's functional suitability through two test series. The results reveal significant shortcomings: The tool's numerical grades and qualitative feedback are often random and do not improve even when its suggestions are incorporated. The highest ratings are achievable only with texts generated by ChatGPT. False claims and nonsensical submissions frequently go undetected, while the implementation of some grading criteria is unreliable and opaque. Since these deficiencies stem from the inherent limitations of large language models (LLMs), fundamental improvements to this or similar tools are not immediately foreseeable. The study critiques the broader trend of adopting AI as a quick fix for systemic problems in education, concluding that Fobizz's marketing of the tool as an objective and time-saving solution is misleading and irresponsible. Finally, the study calls for systematic evaluation and subject-specific pedagogical scrutiny of the use of AI tools in educational contexts.
Authors: Hasan Md Tusfiqur Alam, Devansh Srivastav, Md Abdul Kadir, Daniel Sonntag
Abstract: Deep learning has advanced medical image classification, but interpretability challenges hinder its clinical adoption. This study enhances interpretability in Chest X-ray (CXR) classification by using concept bottleneck models (CBMs) and a multi-agent Retrieval-Augmented Generation (RAG) system for report generation. By modeling relationships between visual features and clinical concepts, we create interpretable concept vectors that guide a multi-agent RAG system to generate radiology reports, enhancing clinical relevance, explainability, and transparency. Evaluation of the generated reports using an LLM-as-a-judge confirmed the interpretability and clinical utility of our model's outputs. On the COVID-QU dataset, our model achieved 81% classification accuracy and demonstrated robust report generation performance, with five key metrics ranging between 84% and 90%. This interpretable multi-agent framework bridges the gap between high-performance AI and the explainability required for reliable AI-driven CXR analysis in clinical settings. Our code is available at https://github.com/tifat58/IRR-with-CBM-RAG.git.
Authors: Tara Radvand, Mojtaba Abdolmaleki, Mohamed Mostagir, Ambuj Tewari
Abstract: Verifying the provenance of content is crucial to the function of many organizations, e.g., educational institutions, social media platforms, firms, etc. This problem is becoming increasingly difficult as text generated by Large Language Models (LLMs) becomes almost indistinguishable from human-generated content. In addition, many institutions utilize in-house LLMs and want to ensure that external, non-sanctioned LLMs do not produce content within the institution. In this paper, we answer the following question: Given a piece of text, can we identify whether it was produced by LLM $A$ or $B$ (where $B$ can be a human)? We model LLM-generated text as a sequential stochastic process with complete dependence on history and design zero-shot statistical tests to distinguish between (i) the text generated by two different sets of LLMs $A$ (in-house) and $B$ (non-sanctioned) and also (ii) LLM-generated and human-generated texts. We prove that the type I and type II errors for our tests decrease exponentially in the text length. In designing our tests, we derive concentration inequalities on the difference between log-perplexity and the average entropy of the string under $A$. Specifically, for a given string, we demonstrate that if the string is generated by $A$, the log-perplexity of the string under $A$ converges to the average entropy of the string under $A$, except with an exponentially small probability in string length. We also show that if $B$ generates the text, except with an exponentially small probability in string length, the log-perplexity of the string under $A$ converges to the average cross-entropy of $B$ and $A$. Lastly, we present preliminary experimental results to support our theoretical results. By enabling guaranteed (with high probability) finding of the origin of harmful LLM-generated text with arbitrary size, we can help combat misinformation.
Authors: Maciej Besta, Julia Barth, Eric Schreiber, Ales Kubicek, Afonso Catarino, Robert Gerstenberger, Piotr Nyczyk, Patrick Iff, Yueling Li, Sam Houliston, Tomasz Sternal, Marcin Copik, Grzegorz Kwa\'sniewski, J\"urgen M\"uller, {\L}ukasz Flis, Hannes Eberhard, Hubert Niewiadomski, Torsten Hoefler
Abstract: Reasoning language models (RLMs), also known as Large Reasoning Models (LRMs), such as OpenAI's o1 and o3, DeepSeek-V3, and Alibaba's QwQ, have redefined AI's problem-solving capabilities by extending LLMs with advanced reasoning mechanisms. Yet, their high costs, proprietary nature, and complex architectures - uniquely combining Reinforcement Learning (RL), search heuristics, and LLMs - present accessibility and scalability challenges. To address these, we propose a comprehensive blueprint that organizes RLM components into a modular framework, based on a survey and analysis of all RLM works. This blueprint incorporates diverse reasoning structures (chains, trees, graphs, and nested forms), reasoning strategies (e.g., Monte Carlo Tree Search, Beam Search), RL concepts (policy, value models and others), supervision schemes (Outcome-Based and Process-Based Supervision), and other related concepts (e.g., Test-Time Compute, Retrieval-Augmented Generation, agent tools). We provide detailed mathematical formulations and algorithmic specifications to simplify RLM implementation. By showing how schemes like LLaMA-Berry, QwQ, Journey Learning, and Graph of Thoughts fit as special cases, we demonstrate the blueprint's versatility and unifying potential. To illustrate its utility, we introduce x1, a modular implementation for rapid RLM prototyping and experimentation. Using x1 and a literature review, we provide key insights, such as multi-phase training for policy and value models, and the importance of familiar training distributions. Finally, we discuss scalable RLM cloud deployments and we outline how RLMs can integrate with a broader LLM ecosystem. Our work demystifies RLM construction, democratizes advanced reasoning capabilities, and fosters innovation, aiming to mitigate the gap between "rich AI" and "poor AI" by lowering barriers to RLM development and experimentation.