Authors: Hang Yang, Hao Chen, Hui Guo, Yineng Chen, Ching-Sheng Lin, Shu Hu, Jinrong Hu, Xi Wu, Xin Wang
Abstract: Accurate and efficient question-answering systems are essential for delivering high-quality patient care in the medical field. While Large Language Models (LLMs) have made remarkable strides across various domains, they continue to face significant challenges in medical question answering, particularly in understanding domain-specific terminologies and performing complex reasoning. These limitations undermine their effectiveness in critical medical applications. To address these issues, we propose a novel approach incorporating similar case generation within a multi-agent medical question-answering (MedQA) system. Specifically, we leverage the Llama3.1:70B model, a state-of-the-art LLM, in a multi-agent architecture to enhance performance on the MedQA dataset using zero-shot learning. Our method capitalizes on the model's inherent medical knowledge and reasoning capabilities, eliminating the need for additional training data. Experimental results show substantial performance gains over existing benchmark models, with improvements of 7% in both accuracy and F1-score across various medical QA tasks. Furthermore, we examine the model's interpretability and reliability in addressing complex medical queries. This research not only offers a robust solution for medical question answering but also establishes a foundation for broader applications of LLMs in the medical domain.
Authors: Shreyas Subramanian, Vikram Elango, Mecit Gungor
Abstract: As foundation AI models continue to increase in size, an important question arises - is massive scale the only path forward? This survey of about 160 papers presents a family of Small Language Models (SLMs) in the 1 to 8 billion parameter range that demonstrate smaller models can perform as well, or even outperform large models. We explore task agnostic, general purpose SLMs, task-specific SLMs and techniques to create SLMs that can guide the community to build models while balancing performance, efficiency, scalability and cost. Furthermore we define and characterize SLMs' effective sizes, representing increased capability with respect to LLMs.
Authors: Pouria Rouzrokh, Moein Shariatnia
Abstract: Systematic literature reviews and meta-analyses are essential for synthesizing research insights, but they remain time-intensive and labor-intensive due to the iterative processes of screening, evaluation, and data extraction. This paper introduces and evaluates LatteReview, a Python-based framework that leverages large language models (LLMs) and multi-agent systems to automate key elements of the systematic review process. Designed to streamline workflows while maintaining rigor, LatteReview utilizes modular agents for tasks such as title and abstract screening, relevance scoring, and structured data extraction. These agents operate within orchestrated workflows, supporting sequential and parallel review rounds, dynamic decision-making, and iterative refinement based on user feedback. LatteReview's architecture integrates LLM providers, enabling compatibility with both cloud-based and locally hosted models. The framework supports features such as Retrieval-Augmented Generation (RAG) for incorporating external context, multimodal reviews, Pydantic-based validation for structured inputs and outputs, and asynchronous programming for handling large-scale datasets. The framework is available on the GitHub repository, with detailed documentation and an installable package.
Authors: Xincheng Wang, Liejun Wang, Yinfeng Yu, Xinxin Jiao
Abstract: Multimodal Sentiment Analysis (MSA) integrates diverse modalities(text, audio, and video) to comprehensively analyze and understand individuals' emotional states. However, the real-world prevalence of incomplete data poses significant challenges to MSA, mainly due to the randomness of modality missing. Moreover, the heterogeneity issue in multimodal data has yet to be effectively addressed. To tackle these challenges, we introduce the Modality-Invariant Bidirectional Temporal Representation Distillation Network (MITR-DNet) for Missing Multimodal Sentiment Analysis. MITR-DNet employs a distillation approach, wherein a complete modality teacher model guides a missing modality student model, ensuring robustness in the presence of modality missing. Simultaneously, we developed the Modality-Invariant Bidirectional Temporal Representation Learning Module (MIB-TRL) to mitigate heterogeneity.
Authors: Liang Xiao, Wen Dai, Shuai Chen, Bin Qin, Chongyang Shi, Haopeng Jing, Tianyu Guo
Abstract: Retrieval-augmented generation has gained significant attention due to its ability to integrate relevant external knowledge, enhancing the accuracy and reliability of the LLMs' responses. Most of the existing methods apply a dynamic multiple retrieval-generating process, to address multi-hop complex questions by decomposing them into sub-problems. However, these methods rely on an unidirectional forward reasoning paradigm, where errors from insufficient reasoning steps or inherent flaws in current retrieval systems are irreversible, potentially derailing the entire reasoning chain. For the first time, this work introduces Retroactive Retrieval-Augmented Generation (RetroRAG), a novel framework to build a retroactive reasoning paradigm. RetroRAG revises and updates the evidence, redirecting the reasoning chain to the correct direction. RetroRAG constructs an evidence-collation-discovery framework to search, generate, and refine credible evidence. It synthesizes inferential evidence related to the key entities in the question from the existing source knowledge and formulates search queries to uncover additional information. As new evidence is found, RetroRAG continually updates and organizes this information, enhancing its ability to locate further necessary evidence. Paired with an Answerer to generate and evaluate outputs, RetroRAG is capable of refining its reasoning process iteratively until a reliable answer is obtained. Empirical evaluations show that RetroRAG significantly outperforms existing methods.
Authors: Mohammad AL-Smadi
Abstract: Recent research has investigated the problem of detecting machine-generated essays for academic purposes. To address this challenge, this research utilizes pre-trained, transformer-based models fine-tuned on Arabic and English academic essays with stylometric features. Custom models based on ELECTRA for English and AraELECTRA for Arabic were trained and evaluated using a benchmark dataset. Proposed models achieved excellent results with an F1-score of 99.7%, ranking 2nd among of 26 teams in the English subtask, and 98.4%, finishing 1st out of 23 teams in the Arabic one.
Authors: Malak Mansour, Ahmed Aly, Bahey Tharwat, Sarim Hashmi, Dong An, Ian Reid
Abstract: Large Language Models (LLMs) such as GPT-4, trained on huge amount of datasets spanning multiple domains, exhibit significant reasoning, understanding, and planning capabilities across various tasks. This study presents the first-ever work in Arabic language integration within the Vision-and-Language Navigation (VLN) domain in robotics, an area that has been notably underexplored in existing research. We perform a comprehensive evaluation of state-of-the-art multi-lingual Small Language Models (SLMs), including GPT-4o mini, Llama 3 8B, and Phi-3 medium 14B, alongside the Arabic-centric LLM, Jais. Our approach utilizes the NavGPT framework, a pure LLM-based instruction-following navigation agent, to assess the impact of language on navigation reasoning through zero-shot sequential action prediction using the R2R dataset. Through comprehensive experiments, we demonstrate that our framework is capable of high-level planning for navigation tasks when provided with instructions in both English and Arabic. However, certain models struggled with reasoning and planning in the Arabic language due to inherent limitations in their capabilities, sub-optimal performance, and parsing issues. These findings highlight the importance of enhancing planning and reasoning capabilities in language models for effective navigation, emphasizing this as a key area for further development while also unlocking the potential of Arabic-language models for impactful real-world applications.
Authors: John C. Rollman, Bruce Rogers, Hamed Zaribafzadeh, Daniel Buckland, Ursula Rogers, Jennifer Gagnon, Ozanan Meireles, Lindsay Jennings, Jim Bennett, Jennifer Nicholson, Nandan Lad, Linda Cendales, Andreas Seas, Alessandro Martinino, E. Shelley Hwang, Allan D. Kirk
Abstract: Background: Healthcare has many manual processes that can benefit from automation and augmentation with Generative Artificial Intelligence (AI), the medical billing and coding process. However, current foundational Large Language Models (LLMs) perform poorly when tasked with generating accurate International Classification of Diseases, 10th edition, Clinical Modification (ICD-10-CM) and Current Procedural Terminology (CPT) codes. Additionally, there are many security and financial challenges in the application of generative AI to healthcare. We present a strategy for developing generative AI tools in healthcare, specifically for medical billing and coding, that balances accuracy, accessibility, and patient privacy. Methods: We fine tune the PHI-3 Mini and PHI-3 Medium LLMs using institutional data and compare the results against the PHI-3 base model, a PHI-3 RAG application, and GPT-4o. We use the post operative surgical report as input and the patients billing claim the associated ICD-10, CPT, and Modifier codes as the target result. Performance is measured by accuracy of code generation, proportion of invalid codes, and the fidelity of the billing claim format. Results: Both fine-tuned models performed better or as well as GPT-4o. The Phi-3 Medium fine-tuned model showed the best performance (ICD-10 Recall and Precision: 72%, 72%; CPT Recall and Precision: 77%, 79%; Modifier Recall and Precision: 63%, 64%). The Phi-3 Medium fine-tuned model only fabricated 1% of ICD-10 codes and 0.6% of CPT codes generated. Conclusions: Our study shows that a small model that is fine-tuned on domain-specific data for specific tasks using a simple set of open-source tools and minimal technological and monetary requirements performs as well as the larger contemporary consumer models.
Authors: Martina Leocata, Alejandro Moreo, Fabrizio Sebastiani
Abstract: The Questio de aqua et terra is a cosmological treatise traditionally attributed to Dante Alighieri. However, the authenticity of this text is controversial, due to discrepancies with Dante's established works and to the absence of contemporary references. This study investigates the authenticity of the Questio via computational authorship verification (AV), a class of techniques which combine supervised machine learning and stylometry. We build a family of AV systems and assemble a corpus of 330 13th- and 14th-century Latin texts, which we use to comparatively evaluate the AV systems through leave-one-out cross-validation. Our best-performing system achieves high verification accuracy (F1=0.970) despite the heterogeneity of the corpus in terms of textual genre. The key contribution to the accuracy of this system is shown to come from Distributional Random Oversampling (DRO), a technique specially tailored to text classification which is here used for the first time in AV. The application of the AV system to the Questio returns a highly confident prediction concerning its authenticity. These findings contribute to the debate on the authorship of the Questio, and highlight DRO's potential in the application of AV to cultural heritage.
Authors: Ashutosh Singh, Rohitash Chandra
Abstract: During the COVID-19 pandemic, community tensions intensified, fuelling Hinduphobic sentiments and discrimination against individuals of Hindu descent within India and worldwide. Large language models (LLMs) have become prominent in natural language processing (NLP) tasks and social media analysis, enabling longitudinal studies of platforms like X (formerly Twitter) for specific issues during COVID-19. We present an abuse detection and sentiment analysis framework that offers a longitudinal analysis of Hinduphobia on X (Twitter) during and after the COVID-19 pandemic. This framework assesses the prevalence and intensity of Hinduphobic discourse, capturing elements such as derogatory jokes and racist remarks through sentiment analysis and abuse detection from pre-trained and fine-tuned LLMs. Additionally, we curate and publish a "Hinduphobic COVID-19 X (Twitter) Dataset" of 8,000 tweets annotated for Hinduphobic abuse detection, which is used to fine-tune a BERT model, resulting in the development of the Hinduphobic BERT (HP-BERT) model. We then further fine-tune HP-BERT using the SenWave dataset for multi-label sentiment analysis. Our study encompasses approximately 27.4 million tweets from six countries, including Australia, Brazil, India, Indonesia, Japan, and the United Kingdom. Our findings reveal a strong correlation between spikes in COVID-19 cases and surges in Hinduphobic rhetoric, highlighting how political narratives, misinformation, and targeted jokes contributed to communal polarisation. These insights provide valuable guidance for developing strategies to mitigate communal tensions in future crises, both locally and globally. We advocate implementing automated monitoring and removal of such content on social media to curb divisive discourse.
Authors: Prashant Verma
Abstract: Document chunking is a critical task in natural language processing (NLP) that involves dividing a document into meaningful segments. Traditional methods often rely solely on semantic analysis, ignoring the spatial layout of elements, which is crucial for understanding relationships in complex documents. This paper introduces a novel hybrid approach that combines layout structure, semantic analysis, and spatial relationships to enhance the cohesion and accuracy of document chunks. By leveraging bounding box information (bbox) and text embeddings, our method constructs a weighted graph representation of document elements, which is then clustered using spectral clustering. Experimental results demonstrate that this approach outperforms traditional methods, particularly in documents with diverse layouts such as reports, articles, and multi-column designs. The proposed method also ensures that no chunk exceeds a specified token length, making it suitable for use cases where token limits are critical (e.g., language models with input size limitations)
Authors: Hussain Ahmad, Diksha Goel
Abstract: The field of Artificial Intelligence (AI) continues to drive transformative innovations, with significant progress in conversational interfaces, autonomous vehicles, and intelligent content creation. Since the launch of ChatGPT in late 2022, the rise of Generative AI has marked a pivotal era, with the term Large Language Models (LLMs) becoming a ubiquitous part of daily life. LLMs have demonstrated exceptional capabilities in tasks such as text summarization, code generation, and creative writing. However, these models are inherently limited by their token-level processing, which restricts their ability to perform abstract reasoning, conceptual understanding, and efficient generation of long-form content. To address these limitations, Meta has introduced Large Concept Models (LCMs), representing a significant shift from traditional token-based frameworks. LCMs use concepts as foundational units of understanding, enabling more sophisticated semantic reasoning and context-aware decision-making. Given the limited academic research on this emerging technology, our study aims to bridge the knowledge gap by collecting, analyzing, and synthesizing existing grey literature to provide a comprehensive understanding of LCMs. Specifically, we (i) identify and describe the features that distinguish LCMs from LLMs, (ii) explore potential applications of LCMs across multiple domains, and (iii) propose future research directions and practical strategies to advance LCM development and adoption.
Authors: Pablo Melendez, Clemens Havas
Abstract: Document layout understanding is a field of study that analyzes the spatial arrangement of information in a document hoping to understand its structure and layout. Models such as LayoutLM (and its subsequent iterations) can understand semi-structured documents with SotA results; however, the lack of open semi-structured data is a limitation in itself. While semi-structured data is common in everyday life (balance sheets, purchase orders, receipts), there is a lack of public datasets for training machine learning models for this type of document. In this investigation we propose a method to generate new, synthetic, layout information that can help overcoming this data shortage. According to our results, the proposed method performs better than LayoutTransformer, another popular layout generation method. We also show that, in some scenarios, text classification can improve when supported by bounding box information.
Authors: Sergei Kudriashov, Veronika Zykova, Angelina Stepanova, Yakov Raskind, Eduard Klyshinsky
Abstract: The interpretation of deep learning models is a rapidly growing field, with particular interest in language models. There are various approaches to this task, including training simpler models to replicate neural network predictions and analyzing the latent space of the model. The latter method allows us to not only identify patterns in the model's decision-making process, but also understand the features of its internal structure. In this paper, we analyze the changes in the internal representation of the BERT model when it is trained with additional grammatical modules and data containing new grammatical structures (polypersonality). We find that adding a single grammatical layer causes the model to separate the new and old grammatical systems within itself, improving the overall performance on perplexity metrics.
Authors: Mohamed Taher Alrefaie, Fatty Salem, Nour Eldin Morsy, Nada Samir, Mohamed Medhat Gaber
Abstract: Understanding how large language models (LLMs) grasp the historical context of concepts and their semantic evolution is essential in advancing artificial intelligence and linguistic studies. This study aims to evaluate the capabilities of various LLMs in capturing temporal dynamics of meaning, specifically how they interpret terms across different time periods. We analyze a diverse set of terms from multiple domains, using tailored prompts and measuring responses through both objective metrics (e.g., perplexity and word count) and subjective human expert evaluations. Our comparative analysis includes prominent models like ChatGPT, GPT-4, Claude, Bard, Gemini, and Llama. Findings reveal marked differences in each model's handling of historical context and semantic shifts, highlighting both strengths and limitations in temporal semantic understanding. These insights offer a foundation for refining LLMs to better address the evolving nature of language, with implications for historical text analysis, AI design, and applications in digital humanities.
Authors: Yuri Facanha Bezerra, Li Weigang
Abstract: We introduce LLMQuoter, a lightweight, distillation-based model designed to enhance Retrieval Augmented Generation (RAG) by extracting the most relevant textual evidence for downstream reasoning tasks. Built on the LLaMA-3B architecture and fine-tuned with Low-Rank Adaptation (LoRA) on a 15,000-sample subset of HotpotQA, LLMQuoter adopts a "quote-first-then-answer" strategy, efficiently identifying key quotes before passing curated snippets to reasoning models. This workflow reduces cognitive overhead and outperforms full-context approaches like Retrieval-Augmented Fine-Tuning (RAFT), achieving over 20-point accuracy gains across both small and large language models. By leveraging knowledge distillation from a high-performing teacher model, LLMQuoter achieves competitive results in a resource-efficient fine-tuning setup. It democratizes advanced RAG capabilities, delivering significant performance improvements without requiring extensive model retraining. Our results highlight the potential of distilled quote-based reasoning to streamline complex workflows, offering a scalable and practical solution for researchers and practitioners alike.
Authors: Adrian Marius Dumitran, Adrian-Catalin Badea, Stefan-Gabriel Muscalu, Angela-Liliana Dumitran, Stefan-Cosmin Dascalescu, Radu-Sebastian Amarie
Abstract: Recent studies have suggested that large language models (LLMs) underperform on mathematical and computer science tasks when these problems are translated from Romanian into English, compared to their original Romanian format. Accurate translation is critical for applications ranging from automatic translations in programming competitions to the creation of high-quality educational materials, as well as minimizing errors or fraud in human translations. This study shows that robust large language models (LLMs) can maintain or even enhance their performance in translating less common languages when given well-structured prompts. Our findings suggest that LLMs, with appropriate supervision, can be reliably used for the automatic translation of IOI (International Olympiad in Informatics)-style tasks. We evaluate several translation methods across multiple LLMs, including OpenRoLLM, Llama 3.1 8B, Llama 3.2 3B and GPT-4o, assessing their translation accuracy and performance stability through repeated runs. Additionally, we augment the OJI (Romanian County-Level Informatics Olympiad) Romanian dataset with accurate English translations, enhancing its utility for future LLM training and evaluation. Through detailed syntactic and semantic analyses, we confirm that with human oversight, LLMs can serve as a viable solution for multilingual problem-solving. We also compare the translation quality of LLMs against human translators, as evaluated by a certified expert, underscoring the potential of LLMs in realworld scenarios.
Authors: Zixuan Liang
Abstract: This paper addresses the harmonization of metadata from diverse repositories of language resources (LRs). Leveraging linked data and RDF techniques, we integrate data from multiple sources into a unified model based on DCAT and META-SHARE OWL ontology. Our methodology supports text-based search, faceted browsing, and advanced SPARQL queries through Linghub, a newly developed portal. Real user queries from the Corpora Mailing List (CML) were evaluated to assess Linghub capability to satisfy actual user needs. Results indicate that while some limitations persist, many user requests can be successfully addressed. The study highlights significant metadata issues and advocates for adherence to open vocabularies and standards to enhance metadata harmonization. This initial research underscores the importance of API-based access to LRs, promoting machine usability and data subset extraction for specific purposes, paving the way for more efficient and standardized LR utilization.
Authors: Rhitabrat Pokharel, Sina Bagheri Nezhad, Ameeta Agrawal, Suresh Singh
Abstract: The rapid advancement of Large Language Models (LLMs), particularly those trained on multilingual corpora, has intensified the need for a deeper understanding of their performance across a diverse range of languages and model sizes. Our research addresses this critical need by studying the performance and scaling behavior of multilingual LLMs in text classification and machine translation tasks across 204 languages. We systematically examine both seen and unseen languages across three model families of varying sizes in zero-shot and few-shot settings. Our findings show significant differences in scaling behavior between zero-shot and two-shot scenarios, with striking disparities in performance between seen and unseen languages. Model scale has little effect on zero-shot performance, which remains mostly flat. However, in two-shot settings, larger models show clear linear improvements in multilingual text classification. For translation tasks, however, only the instruction-tuned model showed clear benefits from scaling. Our analysis also suggests that overall resource levels, not just the proportions of pretraining languages, are better predictors of model performance, shedding light on what drives multilingual LLM effectiveness.
Authors: Zixuan Liang
Abstract: Data preparation, specifically date parsing, is a significant bottleneck in analytic workflows. To address this, we present two algorithms, one based on minimum entropy and the other on natural language modeling that automatically derive date formats from string data. These algorithms achieve over 90% accuracy on a large corpus of data columns, streamlining the data preparation process within visualization environments. The minimal entropy approach is particularly fast, providing interactive feedback. Our methods simplify date format extraction, making them suitable for integration into data visualization tools and databases.
Authors: Anna Marklov\'a, Ji\v{r}\'i Mili\v{c}ka, Leonid Ryvkin, \v{L}udmila Lackov\'a Bennet, Libu\v{s}e Korman\'ikov\'a
Abstract: Lexical iconicity, a direct relation between a word's meaning and its form, is an important aspect of every natural language, most commonly manifesting through sound-meaning associations. Since Large language models' (LLMs') access to both meaning and sound of text is only mediated (meaning through textual context, sound through written representation, further complicated by tokenization), we might expect that the encoding of iconicity in LLMs would be either insufficient or significantly different from human processing. This study addresses this hypothesis by having GPT-4 generate highly iconic pseudowords in artificial languages. To verify that these words actually carry iconicity, we had their meanings guessed by Czech and German participants (n=672) and subsequently by LLM-based participants (generated by GPT-4 and Claude 3.5 Sonnet). The results revealed that humans can guess the meanings of pseudowords in the generated iconic language more accurately than words in distant natural languages and that LLM-based participants are even more successful than humans in this task. This core finding is accompanied by several additional analyses concerning the universality of the generated language and the cues that both human and LLM-based participants utilize.
Authors: Zheqi Lv, Wenkai Wang, Jiawei Wang, Shengyu Zhang, Fei Wu
Abstract: Efficient Multimodal Large Language Models (EMLLMs) have rapidly advanced recently. Incorporating Chain-of-Thought (CoT) reasoning and step-by-step self-evaluation has improved their performance. However, limited parameters often hinder EMLLMs from effectively using self-evaluation during inference. Key challenges include synthesizing evaluation data, determining its quantity, optimizing training and inference strategies, and selecting appropriate prompts. To address these issues, we introduce Self-Evaluation Augmented Training (SEAT). SEAT uses more powerful EMLLMs for CoT reasoning, data selection, and evaluation generation, then trains EMLLMs with the synthesized data. However, handling long prompts and maintaining CoT reasoning quality are problematic. Therefore, we propose Cascaded Self-Evaluation Augmented Training (Cas-SEAT), which breaks down lengthy prompts into shorter, task-specific cascaded prompts and reduces costs for resource-limited settings. During data synthesis, we employ open-source 7B-parameter EMLLMs and annotate a small dataset with short prompts. Experiments demonstrate that Cas-SEAT significantly boosts EMLLMs' self-evaluation abilities, improving performance by 19.68%, 55.57%, and 46.79% on the MathVista, Math-V, and We-Math datasets, respectively. Additionally, our Cas-SEAT Dataset serves as a valuable resource for future research in enhancing EMLLM self-evaluation.
Authors: Aloka Fernando, Surangika Ranathunga
Abstract: Multilingual Pre-trained Language models (multiPLMs), trained on the Masked Language Modelling (MLM) objective are commonly being used for cross-lingual tasks such as bitext mining. However, the performance of these models is still suboptimal for low-resource languages (LRLs). To improve the language representation of a given multiPLM, it is possible to further pre-train it. This is known as continual pre-training. Previous research has shown that continual pre-training with MLM and subsequently with Translation Language Modelling (TLM) improves the cross-lingual representation of multiPLMs. However, during masking, both MLM and TLM give equal weight to all tokens in the input sequence, irrespective of the linguistic properties of the tokens. In this paper, we introduce a novel masking strategy, Linguistic Entity Masking (LEM) to be used in the continual pre-training step to further improve the cross-lingual representations of existing multiPLMs. In contrast to MLM and TLM, LEM limits masking to the linguistic entity types nouns, verbs and named entities, which hold a higher prominence in a sentence. Secondly, we limit masking to a single token within the linguistic entity span thus keeping more context, whereas, in MLM and TLM, tokens are masked randomly. We evaluate the effectiveness of LEM using three downstream tasks, namely bitext mining, parallel data curation and code-mixed sentiment analysis using three low-resource language pairs English-Sinhala, English-Tamil, and Sinhala-Tamil. Experiment results show that continually pre-training a multiPLM with LEM outperforms a multiPLM continually pre-trained with MLM+TLM for all three tasks.
Authors: Vighnesh Subramaniam, Yilun Du, Joshua B. Tenenbaum, Antonio Torralba, Shuang Li, Igor Mordatch
Abstract: Large language models (LLMs) have achieved remarkable performance in recent years but are fundamentally limited by the underlying training data. To improve models beyond the training data, recent works have explored how LLMs can be used to generate synthetic data for autonomous self-improvement. However, successive steps of self-improvement can reach a point of diminishing returns. In this work, we propose a complementary approach towards self-improvement where finetuning is applied to a multiagent society of language models. A group of language models, all starting from the same base model, are independently specialized by updating each one using data generated through multiagent interactions among the models. By training each model on independent sets of data, we illustrate how this approach enables specialization across models and diversification over the set of models. As a result, our overall system is able to preserve diverse reasoning chains and autonomously improve over many more rounds of fine-tuning than single-agent self-improvement methods. We quantitatively illustrate the efficacy of the approach across a wide suite of reasoning tasks.
Authors: Guijin Son, Hyunwoo Ko, Dasol Choi
Abstract: We introduce HRMCR (HAE-RAE Multi-Step Commonsense Reasoning), a benchmark designed to evaluate large language models' ability to perform multi-step reasoning in culturally specific contexts, focusing on Korean. The questions are automatically generated via templates and algorithms, requiring LLMs to integrate Korean cultural knowledge into sequential reasoning steps. Consistent with prior observations on emergent abilities, our experiments reveal that models trained on fewer than \(2 \cdot 10^{25}\) training FLOPs struggle to solve any questions, showing near-zero performance. Beyond this threshold, performance improves sharply. State-of-the-art models (e.g., O1) still score under 50\%, underscoring the difficulty of our tasks. Notably, stepwise analysis suggests the observed emergent behavior may stem from compounding errors across multiple steps rather than reflecting a genuinely new capability. We publicly release the benchmark and commit to regularly updating the dataset to prevent contamination.
Authors: Chen Huang, Yang Deng, Wenqiang Lei, Jiancheng Lv, Tat-Seng Chua, Jimmy Xiangji Huang
Abstract: With the advancement of large language models (LLMs), intelligent models have evolved from mere tools to autonomous agents with their own goals and strategies for cooperating with humans. This evolution has birthed a novel paradigm in NLP, i.e., human-model cooperation, that has yielded remarkable progress in numerous NLP tasks in recent years. In this paper, we take the first step to present a thorough review of human-model cooperation, exploring its principles, formalizations, and open challenges. In particular, we introduce a new taxonomy that provides a unified perspective to summarize existing approaches. Also, we discuss potential frontier areas and their corresponding challenges. We regard our work as an entry point, paving the way for more breakthrough research in this regard.
Authors: Zhengyang Tang, Ziniu Li, Zhenyang Xiao, Tian Ding, Ruoyu Sun, Benyou Wang, Dayiheng Liu, Fei Huang, Tianyu Liu, Bowen Yu, Junyang Lin
Abstract: Despite their remarkable performance, the development of Large Language Models (LLMs) faces a critical challenge in scalable oversight: providing effective feedback for tasks where human evaluation is difficult or where LLMs outperform humans. While there is growing interest in using LLMs for critique, current approaches still rely on human annotations or more powerful models, leaving the issue of enhancing critique capabilities without external supervision unresolved. We introduce SCRIT (Self-evolving CRITic), a framework that enables genuine self-evolution of critique abilities. Technically, SCRIT self-improves by training on synthetic data, generated by a contrastive-based self-critic that uses reference solutions for step-by-step critique, and a self-validation mechanism that ensures critique quality through correction outcomes. Implemented with Qwen2.5-72B-Instruct, one of the most powerful LLMs, SCRIT achieves up to a 10.3\% improvement on critique-correction and error identification benchmarks. Our analysis reveals that SCRIT's performance scales positively with data and model size, outperforms alternative approaches, and benefits critically from its self-validation component.
Authors: Md. Arafat Alam Khandaker, Ziyan Shirin Raha, Bidyarthi Paul, Tashreef Muhammad
Abstract: The Bangla language includes many regional dialects, adding to its cultural richness. The translation of Bangla Language into regional dialects presents a challenge due to significant variations in vocabulary, pronunciation, and sentence structure across regions like Chittagong, Sylhet, Barishal, Noakhali, and Mymensingh. These dialects, though vital to local identities, lack of representation in technological applications. This study addresses this gap by translating standard Bangla into these dialects using neural machine translation (NMT) models, including BanglaT5, mT5, and mBART50. The work is motivated by the need to preserve linguistic diversity and improve communication among dialect speakers. The models were fine-tuned using the "Vashantor" dataset, containing 32,500 sentences across various dialects, and evaluated through Character Error Rate (CER) and Word Error Rate (WER) metrics. BanglaT5 demonstrated superior performance with a CER of 12.3% and WER of 15.7%, highlighting its effectiveness in capturing dialectal nuances. The outcomes of this research contribute to the development of inclusive language technologies that support regional dialects and promote linguistic diversity.
Authors: Hanyu Zhang, Xiting Wang, Chengao Li, Xiang Ao, Qing He
Abstract: As large language models (LLMs) are widely deployed across various domains, the ability to control their generated outputs has become more critical. This control involves aligning LLMs outputs with human values and ethical principles or customizing LLMs on specific topics or styles for individual users. Existing controlled generation methods either require significant computational resources and extensive trial-and-error or provide coarse-grained control. In this paper, we propose Generation with Concept Activation Vector (GCAV), a lightweight model control framework that ensures accurate control without requiring resource-extensive fine-tuning. Specifically, GCAV first trains a concept activation vector for specified concepts to be controlled, such as toxicity. During inference, GCAV steers the concept vector in LLMs, for example, by removing the toxicity concept vector from the activation layers. Control experiments from different perspectives, including toxicity reduction, sentiment control, linguistic style, and topic control, demonstrate that our framework achieves state-of-the-art performance with granular control, allowing for fine-grained adjustments of both the steering layers and the steering magnitudes for individual samples.
Authors: You Li, Heyu Huang, Chi Chen, Kaiyu Huang, Chao Huang, Zonghao Guo, Zhiyuan Liu, Jinan Xu, Yuhua Li, Ruixuan Li, Maosong Sun
Abstract: The recent advancement of Multimodal Large Language Models (MLLMs) has significantly improved their fine-grained perception of single images and general comprehension across multiple images. However, existing MLLMs still face challenges in achieving precise grounding in complex multi-image scenarios. To address this, we first explore a Chain-of-Thought (CoT) framework that integrates single-image grounding with multi-image comprehension. While partially effective, it remains unstable and struggles to capture abstract visual information due to its non-end-to-end nature. Therefore, we introduce Migician, the first multi-image grounding model capable of performing free-form and accurate grounding across multiple images. To support this, we present the MGrounding-630k dataset, which comprises data for several multi-image grounding tasks derived from existing datasets, along with newly generated free-form grounding instruction-following data. Furthermore, we propose MIG-Bench, a comprehensive benchmark specifically designed for evaluating multi-image grounding capabilities. Experimental results demonstrate that our model achieves significantly superior multi-image grounding capabilities, outperforming the best existing MLLMs by 21.61% and even surpassing much larger 70B models. Our code, model, dataset, and benchmark are fully open-sourced.
Authors: Deshan Sumanathilaka, Isuri Anuradha, Ruvan Weerasinghe, Nicholas Micallef, Julian Hough
Abstract: The paper overviews the shared task on Real-Time Reverse Transliteration for Romanized Indo-Aryan languages. It focuses on the reverse transliteration of low-resourced languages in the Indo-Aryan family to their native scripts. Typing Romanized Indo-Aryan languages using ad-hoc transliterals and achieving accurate native scripts are complex and often inaccurate processes with the current keyboard systems. This task aims to introduce and evaluate a real-time reverse transliterator that converts Romanized Indo-Aryan languages to their native scripts, improving the typing experience for users. Out of 11 registered teams, four teams participated in the final evaluation phase with transliteration models for Sinhala, Hindi and Malayalam. These proposed solutions not only solve the issue of ad-hoc transliteration but also empower low-resource language usability in the digital arena.
Authors: Antonin Poch\'e (CERCO, ANITI), Alon Jacovi (CERCO, ANITI), Agustin Martin Picard (CERCO, ANITI), Victor Boutin (CERCO, ANITI), Fanny Jourdan
Abstract: Concept-based explanations work by mapping complex model computations to human-understandable concepts. Evaluating such explanations is very difficult, as it includes not only the quality of the induced space of possible concepts but also how effectively the chosen concepts are communicated to users. Existing evaluation metrics often focus solely on the former, neglecting the latter. We introduce an evaluation framework for measuring concept explanations via automated simulatability: a simulator's ability to predict the explained model's outputs based on the provided explanations. This approach accounts for both the concept space and its interpretation in an end-to-end evaluation. Human studies for simulatability are notoriously difficult to enact, particularly at the scale of a wide, comprehensive empirical evaluation (which is the subject of this work). We propose using large language models (LLMs) as simulators to approximate the evaluation and report various analyses to make such approximations reliable. Our method allows for scalable and consistent evaluation across various models and datasets. We report a comprehensive empirical evaluation using this framework and show that LLMs provide consistent rankings of explanation methods. Code available at https://github.com/AnonymousConSim/ConSim
Authors: Bianca Raimondi, Saverio Giallorenzo, Maurizio Gabbrielli
Abstract: In education, the capability of generating human-like text of Large Language Models (LLMs) inspired work on how they can increase the efficiency of learning and teaching. We study the affordability of these models for educators and students by investigating how LLMs answer multiple-choice questions (MCQs) with respect to hardware constraints and refinement techniques. We explore this space by using generic pre-trained LLMs (the 7B, 13B, and 70B variants of LLaMA-2) to answer 162 undergraduate-level MCQs from a course on Programming Languages (PL) -- the MCQ dataset is a contribution of this work, which we make publicly available. Specifically, we dissect how different factors, such as using readily-available material -- (parts of) the course's textbook -- for fine-tuning and quantisation (to decrease resource usage) can change the accuracy of the responses. The main takeaway is that smaller textbook-based fine-tuned models outperform generic larger ones (whose pre-training requires conspicuous resources), making the usage of LLMs for answering MCQs resource- and material-wise affordable.
Authors: Petraq Nako, Adam Jatowt
Abstract: Predicting future events is an important activity with applications across multiple fields and domains. For example, the capacity to foresee stock market trends, natural disasters, business developments, or political events can facilitate early preventive measures and uncover new opportunities. Multiple diverse computational methods for attempting future predictions, including predictive analysis, time series forecasting, and simulations have been proposed. This study evaluates the performance of several large language models (LLMs) in supporting future prediction tasks, an under-explored domain. We assess the models across three scenarios: Affirmative vs. Likelihood questioning, Reasoning, and Counterfactual analysis. For this, we create a dataset1 by finding and categorizing news articles based on entity type and its popularity. We gather news articles before and after the LLMs training cutoff date in order to thoroughly test and compare model performance. Our research highlights LLMs potential and limitations in predictive modeling, providing a foundation for future improvements.
Authors: Ruby Ostrow, Adam Lopez
Abstract: A large body of research has found substantial gender bias in NLP systems. Most of this research takes a binary, essentialist view of gender: limiting its variation to the categories _men_ and _women_, conflating gender with sex, and ignoring different sexual identities. But gender and sexuality exist on a spectrum, so in this paper we study the biases of large language models (LLMs) towards sexual and gender minorities beyond binary categories. Grounding our study in a widely used psychological framework -- the Stereotype Content Model -- we demonstrate that English-language survey questions about social perceptions elicit more negative stereotypes of sexual and gender minorities from LLMs, just as they do from humans. We then extend this framework to a more realistic use case: text generation. Our analysis shows that LLMs generate stereotyped representations of sexual and gender minorities in this setting, raising concerns about their capacity to amplify representational harms in creative writing, a widely promoted use case.
Authors: Yash Khare, Taufiquzzaman Peyash, Andrea Vanzo, Takuya Yoshioka
Abstract: This paper introduces an all-neural text formatting (TF) model designed for commercial automatic speech recognition (ASR) systems, encompassing punctuation restoration (PR), truecasing, and inverse text normalization (ITN). Unlike traditional rule-based or hybrid approaches, this method leverages a two-stage neural architecture comprising a multi-objective token classifier and a sequence-to-sequence (seq2seq) model. This design minimizes computational costs and reduces hallucinations while ensuring flexibility and robustness across diverse linguistic entities and text domains. Developed as part of the Universal-2 ASR system, the proposed method demonstrates superior performance in TF accuracy, computational efficiency, and perceptual quality, as validated through comprehensive evaluations using both objective and subjective methods. This work underscores the importance of holistic TF models in enhancing ASR usability in practical settings.
Authors: Bennett Kleinberg, Riccardo Loconte, Bruno Verschuere
Abstract: Background: Deception detection through analysing language is a promising avenue using both human judgments and automated machine learning judgments. For both forms of credibility assessment, automated adversarial attacks that rewrite deceptive statements to appear truthful pose a serious threat. Methods: We used a dataset of 243 truthful and 262 fabricated autobiographical stories in a deception detection task for humans and machine learning models. A large language model was tasked to rewrite deceptive statements so that they appear truthful. In Study 1, humans who made a deception judgment or used the detailedness heuristic and two machine learning models (a fine-tuned language model and a simple n-gram model) judged original or adversarial modifications of deceptive statements. In Study 2, we manipulated the target alignment of the modifications, i.e. tailoring the attack to whether the statements would be assessed by humans or computer models. Results: When adversarial modifications were aligned with their target, human (d=-0.07 and d=-0.04) and machine judgments (51% accuracy) dropped to the chance level. When the attack was not aligned with the target, both human heuristics judgments (d=0.30 and d=0.36) and machine learning predictions (63-78%) were significantly better than chance. Conclusions: Easily accessible language models can effectively help anyone fake deception detection efforts both by humans and machine learning models. Robustness against adversarial modifications for humans and machines depends on that target alignment. We close with suggestions on advancing deception research with adversarial attack designs.
Authors: Emil Nuutinen, Iiro Rastas, Filip Ginter
Abstract: We apply a simple method to machine translate datasets with span-level annotation using the DeepL MT service and its ability to translate formatted documents. Using this method, we produce a Finnish version of the SQuAD2.0 question answering dataset and train QA retriever models on this new dataset. We evaluate the quality of the dataset and more generally the MT method through direct evaluation, indirect comparison to other similar datasets, a backtranslation experiment, as well as through the performance of downstream trained QA models. In all these evaluations, we find that the method of transfer is not only simple to use but produces consistently better translated data. Given its good performance on the SQuAD dataset, it is likely the method can be used to translate other similar span-annotated datasets for other tasks and languages as well. All code and data is available under an open license: data at HuggingFace TurkuNLP/squad_v2_fi, code on GitHub TurkuNLP/squad2-fi, and model at HuggingFace TurkuNLP/bert-base-finnish-cased-squad2.
Authors: Eunjung Cho, Won Ik Cho, Soomin Seo
Abstract: Hallucination in large language models (LLMs) remains a significant challenge for their safe deployment, particularly due to its potential to spread misinformation. Most existing solutions address this challenge by focusing on aligning the models with credible sources or by improving how models communicate their confidence (or lack thereof) in their outputs. While these measures may be effective in most contexts, they may fall short in scenarios requiring more nuanced approaches, especially in situations where access to accurate data is limited or determining credible sources is challenging. In this study, we take North Korea - a country characterised by an extreme lack of reliable sources and the prevalence of sensationalist falsehoods - as a case study. We explore and evaluate how some of the best-performing multilingual LLMs and specific language-based models generate information about North Korea in three languages spoken in countries with significant geo-political interests: English (United States, United Kingdom), Korean (South Korea), and Mandarin Chinese (China). Our findings reveal significant differences, suggesting that the choice of model and language can lead to vastly different understandings of North Korea, which has important implications given the global security challenges the country poses.
Authors: Shubham Bansal, Vikas Joshi, Harveen Chadha, Rupeshkumar Mehta, Jinyu Li
Abstract: This study addresses the issue of speaker gender bias in Speech Translation (ST) systems, which can lead to offensive and inaccurate translations. The masculine bias often found in large-scale ST systems is typically perpetuated through training data derived from Machine Translation (MT) systems. Our approach involves two key steps. First, we employ Large Language Models (LLMs) to rectify translations based on the speaker's gender in a cost-effective manner. Second, we fine-tune the ST model with the corrected data, enabling the model to generate gender-specific translations directly from audio cues, without the need for explicit gender input. Additionally, we propose a three-mode fine-tuned model for scenarios where the speaker's gender is either predefined or should not be inferred from speech cues. We demonstrate a 70% improvement in translations for female speakers compared to our baseline and other large-scale ST systems, such as Seamless M4T and Canary, on the MuST-SHE test set.
Authors: Flavio Pisciotta, Ludovica Pannitto, Lucia Busso, Beatrice Bernasconi, Francesca Masini
Abstract: The paper discusses the role of WordNet-based semantic classification in the formalization of constructions, and more specifically in the semantic annotation of schematic fillers, in the Italian Constructicon. We outline how the Italian Constructicon project uses Open Multilingual WordNet topics to represent semantic features and constraints of constructions.
Authors: Romina Oji, Jenny Kunz
Abstract: This paper investigates the optimal use of the multilingual encoder model mDeBERTa for tasks in three Germanic languages -- German, Swedish, and Icelandic -- representing varying levels of presence and likely data quality in mDeBERTas pre-training data. We compare full fine-tuning with the parameter-efficient fine-tuning (PEFT) methods LoRA and Pfeiffer bottleneck adapters, finding that PEFT is more effective for the higher-resource language, German. However, results for Swedish and Icelandic are less consistent. We also observe differences between tasks: While PEFT tends to work better for question answering, full fine-tuning is preferable for named entity recognition. Inspired by previous research on modular approaches that combine task and language adapters, we evaluate the impact of adding PEFT modules trained on unstructured text, finding that this approach is not beneficial.
Authors: Shucong Zhang, Titouan Parcollet, Rogier van Dalen, Sourav Bhattacharya
Abstract: Rotary Position Embedding (RoPE) encodes relative and absolute positional information in Transformer-based models through rotation matrices applied to input vectors within sequences. While RoPE has demonstrated superior performance compared to other positional embedding technologies in natural language processing tasks, its effectiveness in speech processing applications remains understudied. In this work, we conduct a comprehensive evaluation of RoPE across diverse automatic speech recognition (ASR) tasks. Our experimental results demonstrate that for ASR tasks, RoPE consistently achieves lower error rates compared to the currently widely used relative positional embedding. To facilitate further research, we release the implementation and all experimental recipes through the SpeechBrain toolkit.
Authors: Elham Aghakhani, Lu Wang, Karla T. Washington, George Demiris, Jina Huh-Yoo, Rezvaneh Rezapour
Abstract: Problem-solving therapy (PST) is a structured psychological approach that helps individuals manage stress and resolve personal issues by guiding them through problem identification, solution brainstorming, decision-making, and outcome evaluation. As mental health care increasingly integrates technologies like chatbots and large language models (LLMs), understanding how PST can be effectively automated is important. This study leverages anonymized therapy transcripts to analyze and classify therapeutic interventions using various LLMs and transformer-based models. Our results show that GPT-4o achieved the highest accuracy (0.76) in identifying PST strategies, outperforming other models. Additionally, we introduced a new dimension of communication strategies that enhances the current PST framework, offering deeper insights into therapist-client interactions. This research demonstrates the potential of LLMs to automate complex therapeutic dialogue analysis, providing a scalable, efficient tool for mental health interventions. Our annotation framework can enhance the accessibility, effectiveness, and personalization of PST, supporting therapists in real-time with more precise, targeted interventions.
Authors: Fabian David Schmidt, Ivan Vuli\'c, Goran Glava\v{s}, David Ifeoluwa Adelani
Abstract: While recent multilingual automatic speech recognition models claim to support thousands of languages, ASR for low-resource languages remains highly unreliable due to limited bimodal speech and text training data. Better multilingual spoken language understanding (SLU) can strengthen massively the robustness of multilingual ASR by levering language semantics to compensate for scarce training data, such as disambiguating utterances via context or exploiting semantic similarities across languages. Even more so, SLU is indispensable for inclusive speech technology in roughly half of all living languages that lack a formal writing system. However, the evaluation of multilingual SLU remains limited to shallower tasks such as intent classification or language identification. To address this, we present Fleurs-SLU, a multilingual SLU benchmark that encompasses topical speech classification in 102 languages and multiple-choice question answering through listening comprehension in 92 languages. We extensively evaluate both end-to-end speech classification models and cascaded systems that combine speech-to-text transcription with subsequent classification by large language models on Fleurs-SLU. Our results show that cascaded systems exhibit greater robustness in multilingual SLU tasks, though speech encoders can achieve competitive performance in topical speech classification when appropriately pre-trained. We further find a strong correlation between robust multilingual ASR, effective speech-to-text translation, and strong multilingual SLU, highlighting the mutual benefits between acoustic and semantic speech representations.
Authors: Neha Verma, Kenton Murray, Kevin Duh
Abstract: With the rise and ubiquity of larger deep learning models, the need for high-quality compression techniques is growing in order to deploy these models widely. The sheer parameter count of these models makes it difficult to fit them into the memory constraints of different hardware. In this work, we present a novel approach to model compression by merging similar parameter groups within a model, rather than pruning away less important parameters. Specifically, we select, align, and merge separate feed-forward sublayers in Transformer models, and test our method on language modeling, image classification, and machine translation. With our method, we demonstrate performance comparable to the original models while combining more than a third of model feed-forward sublayers, and demonstrate improved performance over a strong layer-pruning baseline. For instance, we can remove over 21% of total parameters from a Vision Transformer, while maintaining 99% of its original performance. Additionally, we observe that some groups of feed-forward sublayers exhibit high activation similarity, which may help explain their surprising mergeability.
Authors: Yuya Asano, Sabit Hassan, Paras Sharma, Anthony Sicilia, Katherine Atwell, Diane Litman, Malihe Alikhani
Abstract: General-purpose automatic speech recognition (ASR) systems do not always perform well in goal-oriented dialogue. Existing ASR correction methods rely on prior user data or named entities. We extend correction to tasks that have no prior user data and exhibit linguistic flexibility such as lexical and syntactic variations. We propose a novel context augmentation with a large language model and a ranking strategy that incorporates contextual information from the dialogue states of a goal-oriented conversational AI and its tasks. Our method ranks (1) n-best ASR hypotheses by their lexical and semantic similarity with context and (2) context by phonetic correspondence with ASR hypotheses. Evaluated in home improvement and cooking domains with real-world users, our method improves recall and F1 of correction by 34% and 16%, respectively, while maintaining precision and false positive rate. Users rated .8-1 point (out of 5) higher when our correction method worked properly, with no decrease due to false positives.
Authors: Xiaodi Li, Dingcheng Li, Rujun Gao, Mahmoud Zamani, Latifur Khan
Abstract: Continual learning has become essential in many practical applications such as online news summaries and product classification. The primary challenge is known as catastrophic forgetting, a phenomenon where a model inadvertently discards previously learned knowledge when it is trained on new tasks. Existing solutions involve storing exemplars from previous classes, regularizing parameters during the fine-tuning process, or assigning different model parameters to each task. The proposed solution LSEBMCL (Latent Space Energy-Based Model for Continual Learning) in this work is to use energy-based models (EBMs) to prevent catastrophic forgetting by sampling data points from previous tasks when training on new ones. The EBM is a machine learning model that associates an energy value with each input data point. The proposed method uses an EBM layer as an outer-generator in the continual learning framework for NLP tasks. The study demonstrates the efficacy of EBM in NLP tasks, achieving state-of-the-art results in all experiments.
Authors: Zheqi Lv, Tianyu Zhan, Wenjie Wang, Xinyu Lin, Shengyu Zhang, Wenqiao Zhang, Jiwei Li, Kun Kuang, Fei Wu
Abstract: Large Language Models (LLMs) for Recommendation (LLM4Rec) is a promising research direction that has demonstrated exceptional performance in this field. However, its inability to capture real-time user preferences greatly limits the practical application of LLM4Rec because (i) LLMs are costly to train and infer frequently, and (ii) LLMs struggle to access real-time data (its large number of parameters poses an obstacle to deployment on devices). Fortunately, small recommendation models (SRMs) can effectively supplement these shortcomings of LLM4Rec diagrams by consuming minimal resources for frequent training and inference, and by conveniently accessing real-time data on devices. In light of this, we designed the Device-Cloud LLM-SRM Collaborative Recommendation Framework (LSC4Rec) under a device-cloud collaboration setting. LSC4Rec aims to integrate the advantages of both LLMs and SRMs, as well as the benefits of cloud and edge computing, achieving a complementary synergy. We enhance the practicability of LSC4Rec by designing three strategies: collaborative training, collaborative inference, and intelligent request. During training, LLM generates candidate lists to enhance the ranking ability of SRM in collaborative scenarios and enables SRM to update adaptively to capture real-time user interests. During inference, LLM and SRM are deployed on the cloud and on the device, respectively. LLM generates candidate lists and initial ranking results based on user behavior, and SRM get reranking results based on the candidate list, with final results integrating both LLM's and SRM's scores. The device determines whether a new candidate list is needed by comparing the consistency of the LLM's and SRM's sorted lists. Our comprehensive and extensive experimental analysis validates the effectiveness of each strategy in LSC4Rec.
Authors: Daowan Peng, Wei Wei
Abstract: Previous studies have pointed out that visual question answering (VQA) models are prone to relying on language priors for answer predictions. In this context, predictions often depend on linguistic shortcuts rather than a comprehensive grasp of multimodal knowledge, which diminishes their generalization ability. In this paper, we propose a novel method, namely, KDAR, leveraging knowledge distillation to address the prior-dependency dilemmas within the VQA task. Specifically, the regularization effect facilitated by soft labels from a well-trained teacher is employed to penalize overfitting to the most common answers. The soft labels, which serve a regularization role, also provide semantic guidance that narrows the range of candidate answers. Additionally, we design an adaptive sample-wise reweighting learning strategy to further mitigate bias by dynamically adjusting the importance of each sample. Experimental results demonstrate that our method enhances performance in both OOD and IID settings. Our method achieves state-of-the-art performance on the VQA-CPv2 out-of-distribution (OOD) benchmark, significantly outperforming previous state-of-the-art approaches.
Authors: Sungjae Lee, Hyejin Park, Jaechang Kim, Jungseul Ok
Abstract: Recent advancements in large language models (LLMs) have shown remarkable potential in various complex tasks requiring multi-step reasoning methods like tree search to explore diverse reasoning paths. However, existing methods often suffer from computational inefficiency and redundancy. First, they overlook the diversity of task difficulties, leading to unnecessarily extensive searches even for easy tasks. Second, they neglect the semantics of reasoning paths, resulting in redundant exploration of semantically identical paths. To address these limitations, we propose Semantic Exploration with Adaptive Gating (SEAG), a computationally efficient method. SEAG employs an adaptive gating mechanism that dynamically decides whether to conduct a tree search, based on the confidence level of answers from a preceding simple reasoning method. Furthermore, its tree-based exploration consolidates semantically identical reasoning steps, reducing redundant explorations while maintaining or even improving accuracy. Our extensive experiments demonstrate that SEAG significantly improves accuracy by 4.3% on average while requiring only 31% of computational costs compared to existing tree search-based methods on complex reasoning benchmarks including GSM8K and ARC with diverse language models such as Llama2, Llama3, and Mistral.
Authors: Matthew Baas, Pieter Scholtz, Arnav Mehta, Elliott Dyson, Akshat Prakash, Herman Kamper
Abstract: Codec-based text-to-speech (TTS) models have shown impressive quality with zero-shot voice cloning abilities. However, they often struggle with more expressive references or complex text inputs. We present MARS6, a robust encoder-decoder transformer for rapid, expressive TTS. MARS6 is built on recent improvements in spoken language modelling. Utilizing a hierarchical setup for its decoder, new speech tokens are processed at a rate of only 12 Hz, enabling efficient modelling of long-form text while retaining reconstruction quality. We combine several recent training and inference techniques to reduce repetitive generation and improve output stability and quality. This enables the 70M-parameter MARS6 to achieve similar performance to models many times larger. We show this in objective and subjective evaluations, comparing TTS output quality and reference speaker cloning ability. Project page: https://camb-ai.github.io/mars6-turbo/
Authors: Soyeong Jeong, Kangsan Kim, Jinheon Baek, Sung Ju Hwang
Abstract: Retrieval-Augmented Generation (RAG) is a powerful strategy to address the issue of generating factually incorrect outputs in foundation models by retrieving external knowledge relevant to queries and incorporating it into their generation process. However, existing RAG approaches have primarily focused on textual information, with some recent advancements beginning to consider images, and they largely overlook videos, a rich source of multimodal knowledge capable of representing events, processes, and contextual details more effectively than any other modality. While a few recent studies explore the integration of videos in the response generation process, they either predefine query-associated videos without retrieving them according to queries, or convert videos into the textual descriptions without harnessing their multimodal richness. To tackle these, we introduce VideoRAG, a novel framework that not only dynamically retrieves relevant videos based on their relevance with queries but also utilizes both visual and textual information of videos in the output generation. Further, to operationalize this, our method revolves around the recent advance of Large Video Language Models (LVLMs), which enable the direct processing of video content to represent it for retrieval and seamless integration of the retrieved videos jointly with queries. We experimentally validate the effectiveness of VideoRAG, showcasing that it is superior to relevant baselines.
Authors: Hongyuan Dong, Zijian Kang, Weijie Yin, Xiao Liang, Chao Feng, Jiao Ran
Abstract: In this paper, we introduce SAIL-VL (ScAlable Vision Language Model TraIning via High QuaLity Data Curation), an open-source vision language model (VLM) of state-of-the-art (SOTA) performance with 2B parameters. We introduce three key improvements that contribute to SAIL-VL's leading performance: (1) Scalable high-quality visual understanding data construction: We implement a visual understanding data construction pipeline, which enables hundred-million-scale high-quality recaption data annotation. Equipped with this pipeline, we curate SAIL-Caption, a large-scale caption dataset with large quantity and the highest data quality compared with opensource caption datasets. (2) Scalable Pretraining with High-Quality Visual Understanding Data: We scale SAIL-VL's pretraining budget up to 131B tokens and show that even a 2B VLM benefits from scaled up training data sizes, exhibiting expected data size scaling laws in visual understanding and instruction following performance. (3) Scalable SFT via quantity and quality scaling: We introduce general guidance for instruction data curation to scale up instruction data continuously, allowing us to construct a large SFT dataset with the highest quality. To further improve SAIL-VL's performance, we propose quality scaling, a multi-stage training recipe with curriculum learning, to improve model performance scaling curves w.r.t. data sizes from logarithmic to be near-linear. SAIL-VL obtains the highest average score in 19 commonly used benchmarks in our evaluation and achieves top1 performance among VLMs of comparable sizes on OpenCompass (https://rank.opencompass.org.cn/leaderboard-multimodal). We release our SAIL-VL-2B model at HuggingFace (https://huggingface.co/BytedanceDouyinContent/SAIL-VL-2B).
URLs: https://rank.opencompass.org.cn/leaderboard-multimodal)., https://huggingface.co/BytedanceDouyinContent/SAIL-VL-2B).
Authors: Ryan Whetten, Lucas Maison, Titouan Parcollet, Marco Dinarelli, Yannick Est\`eve
Abstract: In Self-Supervised Learning (SSL), pre-training and evaluation are resource intensive. In the speech domain, current indicators of the quality of SSL models during pre-training, such as the loss, do not correlate well with downstream performance. Consequently, it is often difficult to gauge the final downstream performance in a cost efficient manner during pre-training. In this work, we propose unsupervised efficient methods that give insights into the quality of the pre-training of SSL speech models, namely, measuring the cluster quality and rank of the embeddings of the SSL model. Results show that measures of cluster quality and rank correlate better with downstream performance than the pre-training loss with only one hour of unlabeled audio, reducing the need for GPU hours and labeled data in SSL model evaluation.
Authors: Charles Goddard, Shamane Siriwardhana, Malikeh Ehghaghi, Luke Meyers, Vlad Karpukhin, Brian Benedict, Mark McQuade, Jacob Solawetz
Abstract: The rapid expansion of the open-source language model landscape presents an opportunity to merge the competencies of these model checkpoints by combining their parameters. Advances in transfer learning, the process of fine-tuning pretrained models for specific tasks, has resulted in the development of vast amounts of task-specific models, typically specialized in individual tasks and unable to utilize each other's strengths. Model merging facilitates the creation of multitask models without the need for additional training, offering a promising avenue for enhancing model performance and versatility. By preserving the intrinsic capabilities of the original models, model merging addresses complex challenges in AI - including the difficulties of catastrophic forgetting and multitask learning. To support this expanding area of research, we introduce MergeKit, a comprehensive, open-source library designed to facilitate the application of model merging strategies. MergeKit offers an extensible framework to efficiently merge models on any hardware, providing utility to researchers and practitioners. To date, thousands of models have been merged by the open-source community, leading to the creation of some of the worlds most powerful open-source model checkpoints, as assessed by the Open LLM Leaderboard. The library is accessible at https://github.com/arcee-ai/MergeKit.
Authors: Paul R\"ottger, Fabio Pernisi, Bertie Vidgen, Dirk Hovy
Abstract: The last two years have seen a rapid growth in concerns around the safety of large language models (LLMs). Researchers and practitioners have met these concerns by creating an abundance of datasets for evaluating and improving LLM safety. However, much of this work has happened in parallel, and with very different goals in mind, ranging from the mitigation of near-term risks around bias and toxic content generation to the assessment of longer-term catastrophic risk potential. This makes it difficult for researchers and practitioners to find the most relevant datasets for their use case, and to identify gaps in dataset coverage that future work may fill. To remedy these issues, we conduct a first systematic review of open datasets for evaluating and improving LLM safety. We review 144 datasets, which we identified through an iterative and community-driven process over the course of several months. We highlight patterns and trends, such as a trend towards fully synthetic datasets, as well as gaps in dataset coverage, such as a clear lack of non-English and naturalistic datasets. We also examine how LLM safety datasets are used in practice -- in LLM release publications and popular LLM benchmarks -- finding that current evaluation practices are highly idiosyncratic and make use of only a small fraction of available datasets. Our contributions are based on SafetyPrompts.com, a living catalogue of open datasets for LLM safety, which we plan to update continuously as the field of LLM safety develops.
Authors: Chankyu Lee, Rajarshi Roy, Mengyao Xu, Jonathan Raiman, Mohammad Shoeybi, Bryan Catanzaro, Wei Ping
Abstract: Decoder-only large language model (LLM)-based embedding models are beginning to outperform BERT or T5-based embedding models in general-purpose text embedding tasks, including dense vector-based retrieval. In this work, we introduce the NV-Embed model, incorporating architectural designs, training procedures, and curated datasets to significantly enhance the performance of LLM as a versatile embedding model, while maintaining its simplicity and reproducibility. For model architecture, we propose a latent attention layer to obtain pooled embeddings, which consistently improves retrieval and downstream task accuracy compared to mean pooling or using the last
Authors: Aryo Pradipta Gema, Joshua Ong Jun Leang, Giwon Hong, Alessio Devoto, Alberto Carlo Maria Mancino, Rohit Saxena, Xuanli He, Yu Zhao, Xiaotang Du, Mohammad Reza Ghasemi Madani, Claire Barale, Robert McHardy, Joshua Harris, Jean Kaddour, Emile van Krieken, Pasquale Minervini
Abstract: Maybe not. We identify and analyse errors in the popular Massive Multitask Language Understanding (MMLU) benchmark. Even though MMLU is widely adopted, our analysis demonstrates numerous ground truth errors that obscure the true capabilities of LLMs. For example, we find that 57% of the analysed questions in the Virology subset contain errors. To address this issue, we introduce a comprehensive framework for identifying dataset errors using a novel error annotation protocol. Then, we create MMLU-Redux, which is a subset of 5,700 manually re-annotated questions across all 57 MMLU subjects. We estimate that 6.49% of MMLU questions contain errors. Using MMLU-Redux, we demonstrate significant discrepancies with the model performance metrics that were originally reported. Our results strongly advocate for revising MMLU's error-ridden questions to enhance its future utility and reliability as a benchmark. https://huggingface.co/datasets/edinburgh-dawg/mmlu-redux-2.0.
URLs: https://huggingface.co/datasets/edinburgh-dawg/mmlu-redux-2.0.
Authors: Mingyang Song, Mao Zheng, Xuan Luo, Yue Pan
Abstract: Utilizing Large Language Models (LLMs) as evaluators to assess the performance of LLMs has garnered attention. However, this kind of evaluation approach is affected by potential biases within LLMs, raising concerns about the accuracy and reliability of the evaluation results of LLMs. To address this problem, we propose and study two many-shot In-Context Learning (ICL) prompt templates to help LLM evaluators mitigate potential biases: Many-Shot with Reference (MSwR) and Many-Shot without Reference (MSoR). Specifically, the former utilizes in-context examples with model-generated evaluation rationales as references, while the latter does not include these references. Using these prompt designs, we investigate the impact of increasing the number of in-context examples on the consistency and quality of the evaluation results. Experimental results show that advanced LLMs, such as GPT-4o, perform better in the many-shot regime than in the zero-shot and few-shot regimes. Furthermore, when using GPT-4o as an evaluator in the many-shot regime, adopting MSwR as the prompt template performs better than MSoR.
Authors: Roman Vashurin, Ekaterina Fadeeva, Artem Vazhentsev, Lyudmila Rvanova, Akim Tsvigun, Daniil Vasilev, Rui Xing, Abdelrahman Boda Sadallah, Kirill Grishchenkov, Sergey Petrakov, Alexander Panchenko, Timothy Baldwin, Preslav Nakov, Maxim Panov, Artem Shelmanov
Abstract: The rapid proliferation of large language models (LLMs) has stimulated researchers to seek effective and efficient approaches to deal with LLM hallucinations and low-quality outputs. Uncertainty quantification (UQ) is a key element of machine learning applications in dealing with such challenges. However, research to date on UQ for LLMs has been fragmented in terms of techniques and evaluation methodologies. In this work, we address this issue by introducing a novel benchmark that implements a collection of state-of-the-art UQ baselines and offers an environment for controllable and consistent evaluation of novel UQ techniques over various text generation tasks. Our benchmark also supports the assessment of confidence normalization methods in terms of their ability to provide interpretable scores. Using our benchmark, we conduct a large-scale empirical investigation of UQ and normalization techniques across eleven tasks, identifying the most effective approaches. Code: https://github.com/IINemo/lm-polygraph Benchmark: https://huggingface.co/LM-Polygraph
URLs: https://github.com/IINemo/lm-polygraph, https://huggingface.co/LM-Polygraph
Authors: Jan Philip Wahle, Terry Ruas, Yang Xu, Bela Gipp
Abstract: Much of the success of modern language models depends on finding a suitable prompt to instruct the model. Until now, it has been largely unknown how variations in the linguistic expression of prompts affect these models. This study systematically and empirically evaluates which linguistic features influence models through paraphrase types, i.e., different linguistic changes at particular positions. We measure behavioral changes for five models across 120 tasks and six families of paraphrases (i.e., morphology, syntax, lexicon, lexico-syntax, discourse, and others). We also control for other prompt engineering factors (e.g., prompt length, lexical diversity, and proximity to training data). Our results show a potential for language models to improve tasks when their prompts are adapted in specific paraphrase types (e.g., 6.7% median gain in Mixtral 8x7B; 5.5% in LLaMA 3 8B). In particular, changes in morphology and lexicon, i.e., the vocabulary used, showed promise in improving prompts. These findings contribute to developing more robust language models capable of handling variability in linguistic expression.
Authors: Mana Makinae, Katsuhito Sudoh, Masaru Yamada, Satoshi Nakamura
Abstract: Simultaneous interpretation (SI), the translation of one language to another in real time, starts translation before the original speech has finished. Its evaluation needs to consider both latency and quality. This trade-off is challenging especially for distant word order language pairs such as English and Japanese. To handle this word order gap, interpreters maintain the word order of the source language as much as possible to keep up with original language to minimize its latency while maintaining its quality, whereas in translation reordering happens to keep fluency in the target language. This means outputs synchronized with the source language are desirable based on the real SI situation, and it's a key for further progress in computational SI and simultaneous machine translation (SiMT). In this work, we propose an automatic evaluation metric for SI and SiMT focusing on word order synchronization. Our evaluation metric is based on rank correlation coefficients, leveraging cross-lingual pre-trained language models. Our experimental results on NAIST-SIC-Aligned and JNPC showed our metrics' effectiveness to measure word order synchronization between source and target language.
Authors: Bowen Wang, Jiuyang Chang, Yiming Qian, Guoxin Chen, Junhao Chen, Zhouqiang Jiang, Jiahao Zhang, Yuta Nakashima, Hajime Nagahara
Abstract: Large language models (LLMs) have recently showcased remarkable capabilities, spanning a wide range of tasks and applications, including those in the medical domain. Models like GPT-4 excel in medical question answering but may face challenges in the lack of interpretability when handling complex tasks in real clinical settings. We thus introduce the diagnostic reasoning dataset for clinical notes (DiReCT), aiming at evaluating the reasoning ability and interpretability of LLMs compared to human doctors. It contains 511 clinical notes, each meticulously annotated by physicians, detailing the diagnostic reasoning process from observations in a clinical note to the final diagnosis. Additionally, a diagnostic knowledge graph is provided to offer essential knowledge for reasoning, which may not be covered in the training data of existing LLMs. Evaluations of leading LLMs on DiReCT bring out a significant gap between their reasoning ability and that of human doctors, highlighting the critical need for models that can reason effectively in real-world clinical scenarios.
Authors: Cl\'ement Dumas, Chris Wendler, Veniamin Veselovsky, Giovanni Monea, Robert West
Abstract: A central question in multilingual language modeling is whether large language models (LLMs) develop a universal concept representation, disentangled from specific languages. In this paper, we address this question by analyzing latent representations (latents) during a word translation task in transformer-based LLMs. We strategically extract latents from a source translation prompt and insert them into the forward pass on a target translation prompt. By doing so, we find that the output language is encoded in the latent at an earlier layer than the concept to be translated. Building on this insight, we conduct two key experiments. First, we demonstrate that we can change the concept without changing the language and vice versa through activation patching alone. Second, we show that patching with the mean over latents across different languages does not impair and instead improves the models' performance in translating the concept. Our results provide evidence for the existence of language-agnostic concept representations within the investigated models.
Authors: Erica Ann Metheney, Lauren Yehle
Abstract: This paper explores and assesses in what ways generative AI can assist in translating survey instruments. Writing effective survey questions is a challenging and complex task, made even more difficult for surveys that will be translated and deployed in multiple linguistic and cultural settings. Translation errors can be detrimental, with known errors rendering data unusable for its intended purpose and undetected errors leading to incorrect conclusions. A growing number of institutions face this problem as surveys deployed by private and academic organizations globalize, and the success of their current efforts depends heavily on researchers' and translators' expertise and the amount of time each party has to contribute to the task. Thus, multilinguistic and multicultural surveys produced by teams with limited expertise, budgets, or time are at significant risk for translation-based errors in their data. We implement a zero-shot prompt experiment using ChatGPT to explore generative AI's ability to identify features of questions that might be difficult to translate to a linguistic audience other than the source language. We find that ChatGPT can provide meaningful feedback on translation issues, including common source survey language, inconsistent conceptualization, sensitivity and formality issues, and nonexistent concepts. In addition, we provide detailed information on the practicality of the approach, including accessing the necessary software, associated costs, and computational run times. Lastly, based on our findings, we propose avenues for future research that integrate AI into survey translation practices.
Authors: Mireia Hernandez Caralt, Ivan Sekuli\'c, Filip Carevi\'c, Nghia Khau, Diana Nicoleta Popa, Bruna Guedes, Victor Guimar\~aes, Zeyu Yang, Andre Manso, Meghana Reddy, Paolo Rosso, Roland Mathis
Abstract: Detecting user frustration in modern-day task-oriented dialog (TOD) systems is imperative for maintaining overall user satisfaction, engagement, and retention. However, most recent research is focused on sentiment and emotion detection in academic settings, thus failing to fully encapsulate implications of real-world user data. To mitigate this gap, in this work, we focus on user frustration in a deployed TOD system, assessing the feasibility of out-of-the-box solutions for user frustration detection. Specifically, we compare the performance of our deployed keyword-based approach, open-source approaches to sentiment analysis, dialog breakdown detection methods, and emerging in-context learning LLM-based detection. Our analysis highlights the limitations of open-source methods for real-world frustration detection, while demonstrating the superior performance of the LLM-based approach, achieving a 16\% relative improvement in F1 score on an internal benchmark. Finally, we analyze advantages and limitations of our methods and provide an insight into user frustration detection task for industry practitioners.
Authors: Rithvik Prakki
Abstract: This paper introduces a novel approach to creating adaptive language agents by integrating active inference with large language models (LLMs). While LLMs demonstrate remarkable capabilities, their reliance on static prompts limits adaptation to new information and changing environments. We address this by implementing an active inference framework that acts as a cognitive layer above an LLM-based agent, dynamically adjusting prompts and search strategies through principled information-seeking behavior. Our framework models the environment using three state factors (prompt, search, and information states) with seven observation modalities capturing quality metrics. By framing the agent's learning through the free energy principle, we enable systematic exploration of prompt combinations and search strategies. Experimental results demonstrate the effectiveness of this approach, with the agent developing accurate models of environment dynamics evidenced by emergent structure in observation matrices. Action selection patterns reveal sophisticated exploration-exploitation behavior, transitioning from initial information-gathering to targeted prompt testing. The integration of thermodynamic principles with language model capabilities provides a principled framework for creating robust, adaptable agents, extending active inference beyond traditional low-dimensional control problems to high-dimensional, language-driven environments.
Authors: Simran Khanuja, Vivek Iyer, Claire He, Graham Neubig
Abstract: Beyond conventional paradigms of translating speech and text, recently, there has been interest in automated transcreation of images to facilitate localization of visual content across different cultures. Attempts to define this as a formal Machine Learning (ML) problem have been impeded by the lack of automatic evaluation mechanisms, with previous work relying solely on human evaluation. In this paper, we seek to close this gap by proposing a suite of automatic evaluation metrics inspired by machine translation (MT) metrics, categorized into: a) Object-based, b) Embedding-based, and c) VLM-based. Drawing on theories from translation studies and real-world transcreation practices, we identify three critical dimensions of image transcreation: cultural relevance, semantic equivalence and visual similarity, and design our metrics to evaluate systems along these axes. Our results show that proprietary VLMs best identify cultural relevance and semantic equivalence, while vision-encoder representations are adept at measuring visual similarity. Meta-evaluation across 7 countries shows our metrics agree strongly with human ratings, with average segment-level correlations ranging from 0.55-0.87. Finally, through a discussion of the merits and demerits of each metric, we offer a robust framework for automated image transcreation evaluation, grounded in both theoretical foundations and practical application. Our code can be found here: https://github.com/simran-khanuja/automatic-eval-transcreation
URLs: https://github.com/simran-khanuja/automatic-eval-transcreation
Authors: Benjamin Reichman, Xiaofan Yu, Lanxiang Hu, Jack Truxal, Atishay Jain, Rushil Chandrupatla, Tajana \v{S}imuni\'c Rosing, Larry Heck
Abstract: With the rapid growth in sensor data, effectively interpreting and interfacing with these data in a human-understandable way has become crucial. While existing research primarily focuses on learning classification models, fewer studies have explored how end users can actively extract useful insights from sensor data, often hindered by the lack of a proper dataset. To address this gap, we introduce SensorQA, the first human-created question-answering (QA) dataset for long-term time-series sensor data for daily life monitoring. SensorQA is created by human workers and includes 5.6K diverse and practical queries that reflect genuine human interests, paired with accurate answers derived from sensor data. We further establish benchmarks for state-of-the-art AI models on this dataset and evaluate their performance on typical edge devices. Our results reveal a gap between current models and optimal QA performance and efficiency, highlighting the need for new contributions. The dataset and code are available at: \url{https://github.com/benjamin-reichman/SensorQA}.
Authors: Peter Baile Chen, Yi Zhang, Dan Roth
Abstract: Retrieving relevant tables containing the necessary information to accurately answer a given question over tables is critical to open-domain question-answering (QA) systems. Previous methods assume the answer to such a question can be found either in a single table or multiple tables identified through question decomposition or rewriting. However, neither of these approaches is sufficient, as many questions require retrieving multiple tables and joining them through a join plan that cannot be discerned from the user query itself. If the join plan is not considered in the retrieval stage, the subsequent steps of reasoning and answering based on those retrieved tables are likely to be incorrect. To address this problem, we introduce a method that uncovers useful join relations for any query and database during table retrieval. We use a novel re-ranking method formulated as a mixed-integer program that considers not only table-query relevance but also table-table relevance that requires inferring join relationships. Our method outperforms the state-of-the-art approaches for table retrieval by up to 9.3% in F1 score and for end-to-end QA by up to 5.4% in accuracy.
Authors: Ridouane Ghermi, Xi Wang, Vicky Kalogeiton, Ivan Laptev
Abstract: Recent developments in vision-language models have significantly advanced video understanding. Existing datasets and tasks, however, have notable limitations. Most datasets are confined to short videos with limited events and narrow narratives. For example, datasets with instructional and egocentric videos often depict activities of one person in a single scene. Although existing movie datasets offer richer content, they are often limited to short-term tasks, lack publicly available videos, and frequently encounter data leakage issues given the use of subtitles and other information about commercial movies during LLM pretraining. To address the above limitations, we propose Short-Films 20K (SF20K), the largest publicly available movie dataset. SF20K is composed of 20,143 amateur films and offers long-term video tasks in the form of multiple-choice and open-ended question answering. Our extensive analysis of SF20K reveals minimal data leakage, emphasizes the need for long-term reasoning, and demonstrates the strong performance of recent VLMs. Finally, we show that instruction tuning on the SF20K-Train set substantially improves model performance, paving the way for future progress in long-term video understanding.
Authors: Jiahai Feng, Stuart Russell, Jacob Steinhardt
Abstract: Pretrained language models (LMs) can generalize to implications of facts that they are finetuned on. For example, if finetuned on ``John Doe lives in Tokyo," LMs can correctly answer ``What language do the people in John Doe's city speak?'' with ``Japanese''. However, little is known about the mechanisms that enable this generalization or how they are learned during pretraining. We introduce extractive structures as a framework for describing how components in LMs (e.g., MLPs or attention heads) coordinate to enable this generalization. The structures consist of informative components that store training facts as weight changes, and upstream and downstream extractive components that query and process the stored information to produce the correct implication. We hypothesize that extractive structures are learned during pretraining when encountering implications of previously known facts. This yields two predictions: a data ordering effect where extractive structures can be learned only if facts precede their implications, and a weight grafting effect where extractive structures can be transferred to predict counterfactual implications. We empirically demonstrate these phenomena in the OLMo-7b, Llama 3-8b, Gemma 2-9b, and Qwen 2-7b models. Of independent interest, our results also indicate that fact learning can occur at both early and late layers, which lead to different forms of generalization.
Authors: Renqiu Xia, Mingsheng Li, Hancheng Ye, Wenjie Wu, Hongbin Zhou, Jiakang Yuan, Tianshuo Peng, Xinyu Cai, Xiangchao Yan, Bin Wang, Conghui He, Botian Shi, Tao Chen, Junchi Yan, Bo Zhang
Abstract: Despite their proficiency in general tasks, Multi-modal Large Language Models (MLLMs) struggle with automatic Geometry Problem Solving (GPS), which demands understanding diagrams, interpreting symbols, and performing complex reasoning. This limitation arises from their pre-training on natural images and texts, along with the lack of automated verification in the problem-solving process. Besides, current geometric specialists are limited by their task-specific designs, making them less effective for broader geometric problems. To this end, we present GeoX, a multi-modal large model focusing on geometric understanding and reasoning tasks. Given the significant differences between geometric diagram-symbol and natural image-text, we introduce unimodal pre-training to develop a diagram encoder and symbol decoder, enhancing the understanding of geometric images and corpora. Furthermore, we introduce geometry-language alignment, an effective pre-training paradigm that bridges the modality gap between unimodal geometric experts. We propose a Generator-And-Sampler Transformer (GS-Former) to generate discriminative queries and eliminate uninformative representations from unevenly distributed geometric signals. Finally, GeoX benefits from visual instruction tuning, empowering it to take geometric images and questions as input and generate verifiable solutions. Experiments show that GeoX outperforms both generalists and geometric specialists on publicly recognized benchmarks, such as GeoQA, UniGeo, Geometry3K, and PGPS9k.
Authors: Daniel Paleka, Abhimanyu Pallavi Sudhir, Alejandro Alvarez, Vineeth Bhat, Adam Shen, Evan Wang, Florian Tram\`er
Abstract: Forecasting is a task that is difficult to evaluate: the ground truth can only be known in the future. Recent work showing LLM forecasters rapidly approaching human-level performance begs the question: how can we benchmark and evaluate these forecasters instantaneously? Following the consistency check framework, we measure the performance of forecasters in terms of the consistency of their predictions on different logically-related questions. We propose a new, general consistency metric based on arbitrage: for example, if a forecasting AI illogically predicts that both the Democratic and Republican parties have 60% probability of winning the 2024 US presidential election, an arbitrageur can trade against the forecaster's predictions and make a profit. We build an automated evaluation system that generates a set of base questions, instantiates consistency checks from these questions, elicits the predictions of the forecaster, and measures the consistency of the predictions. We then build a standard, proper-scoring-rule forecasting benchmark, and show that our (instantaneous) consistency metrics correlate with LLM forecasters' ground truth Brier scores (which are only known in the future). We also release a consistency benchmark that resolves in 2028, providing a long-term evaluation tool for forecasting.
Authors: Zongxia Li, Xiyang Wu, Hongyang Du, Huy Nghiem, Guangyao Shi
Abstract: Multimodal Vision Language Models (VLMs) have emerged as a transformative technology at the intersection of computer vision and natural language processing, enabling machines to perceive and reason about the world through both visual and textual modalities. For example, models such as CLIP, Claude, and GPT-4V demonstrate strong reasoning and understanding abilities on visual and textual data and beat classical single modality vision models on zero-shot classification. Despite their rapid advancements in research and growing popularity in applications, a comprehensive survey of existing studies on VLMs is notably lacking, particularly for researchers aiming to leverage VLMs in their specific domains. To this end, we provide a systematic overview of VLMs in the following aspects: model information of the major VLMs developed over the past five years (2019-2024); the main architectures and training methods of these VLMs; summary and categorization of the popular benchmarks and evaluation metrics of VLMs; the applications of VLMs including embodied agents, robotics, and video generation; the challenges and issues faced by current VLMs such as hallucination, fairness, and safety. Detailed collections including papers and model repository links are listed in https://github.com/zli12321/Awesome-VLM-Papers-And-Models.git.
URLs: https://github.com/zli12321/Awesome-VLM-Papers-And-Models.git.
Authors: Jiakang Yuan, Xiangchao Yan, Botian Shi, Tao Chen, Wanli Ouyang, Bo Zhang, Lei Bai, Yu Qiao, Bowen Zhou
Abstract: The scientific research paradigm is undergoing a profound transformation owing to the development of Artificial Intelligence (AI). Recent works demonstrate that various AI-assisted research methods can largely improve research efficiency by improving data analysis, accelerating computation, and fostering novel idea generation. To further move towards the ultimate goal (i.e., automatic scientific research), in this paper, we propose Dolphin, the first closed-loop open-ended auto-research framework to further build the entire process of human scientific research. Dolphin can generate research ideas, perform experiments, and get feedback from experimental results to generate higher-quality ideas. More specifically, Dolphin first generates novel ideas based on relevant papers which are ranked by the topic and task attributes. Then, the codes are automatically generated and debugged with the exception-traceback-guided local code structure. Finally, Dolphin automatically analyzes the results of each idea and feeds the results back to the next round of idea generation. Experiments are conducted on the benchmark datasets of different topics and results show that Dolphin can generate novel ideas continuously and complete the experiment in a loop. We highlight that Dolphin can automatically propose methods that are comparable to the state-of-the-art in some tasks such as 2D image classification and 3D point classification.