abstracts list | id_1 string | id_2 string | pair_id string |
|---|---|---|---|
[
" The handling of communication breakdowns and loss of engagement is an\nimportant aspect of spoken dialogue systems, particularly for chatting systems\nsuch as attentive listening, where the user is mostly speaking. We presume that\na human is best equipped to handle this task and rescue the flow of\nconversation... | 2402.14863 | 2309.16349 | 2402.14863_2309.16349 |
[
" The main objective of Knowledge Graph (KG) embeddings is to learn\nlow-dimensional representations of entities and relations, enabling the\nprediction of missing facts. A significant challenge in achieving better KG\nembeddings lies in capturing relation patterns, including symmetry,\nantisymmetry, inversion, co... | 2305.13015 | 2404.00599 | 2305.13015_2404.00599 |
[
" Gisting (Mu et al., 2023) is a simple method for training models to compress\ninformation into fewer token representations using a modified attention mask,\nand can serve as an economical approach to training Transformer-based\nhypernetworks. We introduce HyperLlama, a set of Gisting-based hypernetworks\nbuilt o... | 2402.16817 | 2402.03710 | 2402.16817_2402.03710 |
[
" Large language model inference on Central Processing Units (CPU) is\nchallenging due to the vast quantities of expensive Multiply-Add (MAD) matrix\noperations in the attention computations. In this paper, we argue that there is\na rare gem in modern CPUs, Single-Instruction-Multiple-Data (SIMD) registers,\nwhich... | 2403.01273 | 2403.00774 | 2403.01273_2403.00774 |
[
" Existing event-centric NLP models often only apply to the pre-defined\nontology, which significantly restricts their generalization capabilities. This\npaper presents CEO, a novel Corpus-based Event Ontology induction model to\nrelax the restriction imposed by pre-defined event ontologies. Without direct\nsuperv... | 2305.13521 | 2402.14852 | 2305.13521_2402.14852 |
[
" This technical report aims to fill a deficiency in the assessment of large\nmultimodal models (LMMs) by specifically examining the self-consistency of\ntheir outputs when subjected to common corruptions. We investigate the\ncross-modal interactions between text, image, and speech, encompassing four\nessential ge... | 2401.11943 | 2310.08483 | 2401.11943_2310.08483 |
[
" Model explanations can be valuable for interpreting and debugging predictive\nmodels. We study a specific kind called Concept Explanations, where the goal is\nto interpret a model using human-understandable concepts. Although popular for\ntheir easy interpretation, concept explanations are known to be noisy. We ... | 2312.08063 | 2404.02372 | 2312.08063_2404.02372 |
[
" In open-domain Question Answering (QA), dense retrieval is crucial for\nfinding relevant passages for answer generation. Typically, contrastive\nlearning is used to train a retrieval model that maps passages and queries to\nthe same semantic space. The objective is to make similar ones closer and\ndissimilar one... | 2401.00165 | 2401.12406 | 2401.00165_2401.12406 |
[
" Student commitment towards a learning recommendation is not separable from\ntheir understanding of the reasons it was recommended to them; and their\nability to modify it based on that understanding. Among explainability\napproaches, chatbots offer the potential to engage the student in a\nconversation, similar ... | 2401.08517 | 2404.03823 | 2401.08517_2404.03823 |
[
" Few-shot relation extraction involves identifying the type of relationship\nbetween two specific entities within a text, using a limited number of\nannotated samples. A variety of solutions to this problem have emerged by\napplying meta-learning and neural graph techniques which typically necessitate\na training... | 2311.05922 | 2403.04132 | 2311.05922_2403.04132 |
[
" Aligning large language models (LLMs) with human values is a vital task for\nLLM practitioners. Current alignment techniques have several limitations: (1)\nrequiring a large amount of annotated data; (2) demanding heavy human\ninvolvement; (3) lacking a systematic mechanism to continuously improve. In\nthis work... | 2401.06785 | 2403.00823 | 2401.06785_2403.00823 |
[
" This paper reports our work on building up a Cantonese Speech-to-Text (STT)\nsystem with a syllable based acoustic model. This is a part of an effort in\nbuilding a STT system to aid dyslexic students who have cognitive deficiency in\nwriting skills but have no problem expressing their ideas through speech. For\... | 2402.08788 | 2402.09967 | 2402.08788_2402.09967 |
[
" Recent advances in large language models (LLMs) have demonstrated potential\nfor LLM agents. To facilitate the training for these agents with both\nlinguistic feedback and non-linguistic reward signals, we introduce Learning\nthrough Communication (LTC). We design a universal buffer to store all the\nfeedback, a... | 2310.01444 | 2404.10500 | 2310.01444_2404.10500 |
[
" Recently, various merging methods have been proposed to build a multi-task\nmodel from task-specific finetuned models without retraining. However, existing\nmethods suffer from a large performance deterioration compared to using\nmultiple task-specific models. In this paper, we propose to inject\ntask-specific k... | 2310.01886 | 2401.15780 | 2310.01886_2401.15780 |
[
" Large language models (LLMs) are highly effective in various natural language\nprocessing (NLP) tasks. However, they are susceptible to producing unreliable\nconjectures in ambiguous contexts called hallucination. This paper presents a\nnew method for evaluating LLM hallucination in Question Answering (QA) based... | 2403.03558 | 2310.09342 | 2403.03558_2310.09342 |
[
" Automatically generating scripts (i.e. sequences of key steps described in\ntext) from video demonstrations and reasoning about the subsequent steps are\ncrucial to the modern AI virtual assistants to guide humans to complete\neveryday tasks, especially unfamiliar ones. However, current methods for\ngenerative s... | 2310.04965 | 2403.11129 | 2310.04965_2403.11129 |
[
" Multilingual modelling can improve machine translation for low-resource\nlanguages, partly through shared subword representations. This paper studies\nthe role of subword segmentation in cross-lingual transfer. We systematically\ncompare the efficacy of several subword methods in promoting synergy and\npreventin... | 2403.20157 | 2403.15690 | 2403.20157_2403.15690 |
[
" Many reinforcement learning environments (e.g., Minecraft) provide only\nsparse rewards that indicate task completion or failure with binary values. The\nchallenge in exploration efficiency in such environments makes it difficult for\nreinforcement-learning-based agents to learn complex tasks. To address this,\n... | 2312.09238 | 2402.09642 | 2312.09238_2402.09642 |
[
" Despite the considerable efforts being made to monitor and regulate\nuser-generated content on social media platforms, the pervasiveness of\noffensive language, such as hate speech or cyberbullying, in the digital space\nremains a significant challenge. Given the importance of maintaining a\ncivilized and respec... | 2403.18314 | 2402.12150 | 2403.18314_2402.12150 |
[
" While large language models (LLMs) have demonstrated exceptional performance\nin recent natural language processing (NLP) tasks, their deployment poses\nsubstantial challenges due to high computational and memory demands in\nreal-world applications. Recent studies have focused on enhancing smaller\nmodels throug... | 2312.10730 | 2404.10838 | 2312.10730_2404.10838 |
[
" The recent explosion in the capabilities of large language models has led to\na wave of interest in how best to prompt a model to perform a given task. While\nit may be tempting to simply choose a prompt based on average performance on a\nvalidation set, this can lead to a deployment where unexpectedly poor resp... | 2311.13628 | 2404.05225 | 2311.13628_2404.05225 |
[
" The training paradigm for machine translation has gradually shifted, from\nlearning neural machine translation (NMT) models with extensive parallel\ncorpora to instruction finetuning on multilingual large language models (LLMs)\nwith high-quality translation pairs. In this paper, we focus on boosting\nmany-to-ma... | 2401.05861 | 2402.06820 | 2401.05861_2402.06820 |
[
" While large language models (LLMs) often adopt finetuning to unlock their\ncapabilities for downstream applications, our understanding on the inductive\nbiases (especially the scaling properties) of different finetuning methods is\nstill limited. To fill this gap, we conduct systematic experiments studying\nwhet... | 2402.17193 | 2403.20184 | 2402.17193_2403.20184 |
[
" We present an overview of the BLP Sentiment Shared Task, organized as part of\nthe inaugural BLP 2023 workshop, co-located with EMNLP 2023. The task is\ndefined as the detection of sentiment in a given piece of social media text.\nThis task attracted interest from 71 participants, among whom 29 and 30 teams\nsub... | 2310.16183 | 2403.02884 | 2310.16183_2403.02884 |
[
" Guaranteeing the correctness and factuality of language model (LM) outputs is\na major open problem. In this work, we propose conformal factuality, a\nframework that can ensure high probability correctness guarantees for LMs by\nconnecting language modeling and conformal prediction. We observe that the\ncorrectn... | 2402.10978 | 2402.12998 | 2402.10978_2402.12998 |
[
" Sentiment analysis is an important tool for aggregating patient voices, in\norder to provide targeted improvements in healthcare services. A prerequisite\nfor this is the availability of in-domain data annotated for sentiment. This\narticle documents an effort to add sentiment annotations to free-text comments\n... | 2404.18832 | 2305.09955 | 2404.18832_2305.09955 |
[
" The eXtreme Multi-label text Classification(XMC) refers to training a\nclassifier that assigns a text sample with relevant labels from an extremely\nlarge-scale label set (e.g., millions of labels). We propose MatchXML, an\nefficient text-label matching framework for XMC. We observe that the label\nembeddings ge... | 2308.13139 | 2403.15492 | 2308.13139_2403.15492 |
[
" Topic models are a popular tool for understanding text collections, but their\nevaluation has been a point of contention. Automated evaluation metrics such as\ncoherence are often used, however, their validity has been questioned for\nneural topic models (NTMs) and can overlook a models benefits in real world\na... | 2401.16348 | 0808.0521 | 2401.16348_0808.0521 |
[
" How can we compress language models without sacrificing accuracy? The number\nof compression algorithms for language models is rapidly growing to benefit\nfrom remarkable advances of recent language models without side effects due to\nthe gigantic size of language models, such as increased carbon emissions and\n... | 2401.15347 | 2403.19851 | 2401.15347_2403.19851 |
[
" While large language models (LLMs) have demonstrated exceptional performance\nin recent natural language processing (NLP) tasks, their deployment poses\nsubstantial challenges due to high computational and memory demands in\nreal-world applications. Recent studies have focused on enhancing smaller\nmodels throug... | 2312.10730 | 2403.18093 | 2312.10730_2403.18093 |
[
" The use of low-rank adaptation (LoRA) with frozen pretrained language models\n(PLMs) has become increasing popular as a mainstream, resource-efficient\nmodeling approach for memory-constrained hardware. In this study, we first\nexplore how to enhance model performance by introducing various LoRA training\nstrate... | 2401.10447 | 2208.10160 | 2401.10447_2208.10160 |
[
" We investigate automatic interlinear glossing in low-resource settings. We\naugment a hard-attentional neural model with embedded translation information\nextracted from interlinear glossed text. After encoding these translations\nusing large language models, specifically BERT and T5, we introduce a\ncharacter-l... | 2403.08189 | 2311.09889 | 2403.08189_2311.09889 |
[
" The rapid development of Large Language Models (LLMs) has facilitated a\nvariety of applications from different domains. In this technical report, we\nexplore the integration of LLMs and the popular academic writing tool,\nOverleaf, to enhance the efficiency and quality of academic writing. To achieve\nthe above... | 2403.09733 | 2404.11086 | 2403.09733_2404.11086 |
[
" Hypertension, defined as blood pressure (BP) that is above normal, holds\nparamount significance in the realm of public health, as it serves as a\ncritical precursor to various cardiovascular diseases (CVDs) and significantly\ncontributes to elevated mortality rates worldwide. However, many existing BP\nmeasurem... | 2402.01826 | 2308.06463 | 2402.01826_2308.06463 |
[
" Effective diabetes management is crucial for maintaining health in diabetic\npatients. Large Language Models (LLMs) have opened new avenues for diabetes\nmanagement, facilitating their efficacy. However, current LLM-based approaches\nare limited by their dependence on general sources and lack of integration with... | 2402.10153 | 2402.16358 | 2402.10153_2402.16358 |
[
" Antonyms vs synonyms distinction is a core challenge in lexico-semantic\nanalysis and automated lexical resource construction. These pairs share a\nsimilar distributional context which makes it harder to distinguish them.\nLeading research in this regard attempts to capture the properties of the\nrelation pairs,... | 2401.10045 | 2307.12798 | 2401.10045_2307.12798 |
[
" Generative language models are usually pretrained on large text corpus via\npredicting the next token (i.e., sub-word/word/phrase) given the previous ones.\nRecent works have demonstrated the impressive performance of large generative\nlanguage models on downstream tasks. However, existing generative language\nm... | 2310.19531 | 2302.10199 | 2310.19531_2302.10199 |
[
" Reading comprehension continues to be a crucial research focus in the NLP\ncommunity. Recent advances in Machine Reading Comprehension (MRC) have mostly\ncentered on literal comprehension, referring to the surface-level understanding\nof content. In this work, we focus on the next level - interpretive\ncomprehen... | 2404.05250 | 2401.02772 | 2404.05250_2401.02772 |
[
" The task of accurate and efficient language translation is an extremely\nimportant information processing task. Machine learning enabled and automated\ntranslation that is accurate and fast is often a large topic of interest in the\nmachine learning and data science communities. In this study, we examine using\n... | 2404.14680 | 2205.15231 | 2404.14680_2205.15231 |
[
" We investigate two research questions: (1) how do machine translation (MT)\nand diacritization influence the performance of each other in a multi-task\nlearning setting (2) the effect of keeping (vs. removing) diacritics on MT\nperformance. We examine these two questions in both high-resource (HR) and\nlow-resou... | 2404.05943 | 2003.11517 | 2404.05943_2003.11517 |
[
" Since the breakthrough of ChatGPT, large language models (LLMs) have garnered\nsignificant attention in the research community. With the development of LLMs,\nthe question of text style transfer for conversational models has emerged as a\nnatural extension, where chatbots may possess their own styles or even\nch... | 2403.08943 | 2402.11569 | 2403.08943_2402.11569 |
[
" The explanations of large language models have recently been shown to be\nsensitive to the randomness used for their training, creating a need to\ncharacterize this sensitivity. In this paper, we propose a characterization\nthat questions the possibility to provide simple and informative explanations\nfor such m... | 2403.10275 | 2402.13528 | 2403.10275_2402.13528 |
[
" Large Language Models (LLMs) have come closest among all models to date to\nmastering human language, yet opinions about their linguistic and cognitive\ncapabilities remain split. Here, we evaluate LLMs using a distinction between\nformal linguistic competence -- knowledge of linguistic rules and patterns --\nan... | 2301.06627 | 2312.03724 | 2301.06627_2312.03724 |
[
" Discussion and debate among Large Language Models (LLMs) have gained\nconsiderable attention due to their potential to enhance the reasoning ability\nof LLMs. Although natural language is an obvious choice for communication due\nto LLM's language understanding capability, the token sampling step needed when\ngen... | 2310.06272 | 2401.10900 | 2310.06272_2401.10900 |
[
" As an efficient model for knowledge organization, the knowledge graph has\nbeen widely adopted in several fields, e.g., biomedicine, sociology, and\neducation. And there is a steady trend of learning embedding representations of\nknowledge graphs to facilitate knowledge graph construction and downstream\ntasks. ... | 1911.08776 | 2402.02144 | 1911.08776_2402.02144 |
[
" In this paper, we present an innovative process-oriented math process reward\nmodel called \\textbf{Math-Shepherd}, which assigns a reward score to each step\nof math problem solutions. The training of Math-Shepherd is achieved using\nautomatically constructed process-wise supervision data, breaking the\nbottlen... | 2312.08935 | 2402.01698 | 2312.08935_2402.01698 |
[
" Geoparsing is the task of estimating the latitude and longitude (coordinates)\nof location expressions in texts. Geoparsing must deal with the ambiguity of\nthe expressions that indicate multiple locations with the same notation. For\nevaluating geoparsing systems, several corpora have been proposed in previous\... | 2403.16483 | 2404.05632 | 2403.16483_2404.05632 |
[
" In recent years, there has been a growing interest in integrating linear\nstate-space models (SSM) in deep neural network architectures of foundation\nmodels. This is exemplified by the recent success of Mamba, showing better\nperformance than the state-of-the-art Transformer architectures in language\ntasks. Fo... | 2403.16899 | 2403.18018 | 2403.16899_2403.18018 |
[
" The coverage and composition of the pretraining data significantly impacts\nthe generalization ability of Large Language Models (LLMs). Despite its\nimportance, recent LLMs still rely on heuristics and trial and error to\nincrease or reduce the influence of data-domains. We propose DOmain reweighting\nwith Gener... | 2310.15393 | 2404.04204 | 2310.15393_2404.04204 |
[
" Generating coherent and credible explanations remains a significant challenge\nin the field of AI. In recent years, researchers have delved into the\nutilization of entailment trees to depict explanations, which exhibit a\nreasoning process of how a hypothesis is deduced from the supporting facts.\nHowever, exis... | 2403.06410 | 2401.09454 | 2403.06410_2401.09454 |
[
" Recent years have seen the rise of large language models (LLMs), where\npractitioners use task-specific prompts; this was shown to be effective for a\nvariety of tasks. However, when applied to semantic textual similarity (STS)\nand natural language inference (NLI), the effectiveness of LLMs turns out to be\nlim... | 2309.08969 | 2311.03663 | 2309.08969_2311.03663 |
[
" Charts provide visual representations of data and are widely used for\nanalyzing information, addressing queries, and conveying insights to others.\nVarious chart-related downstream tasks have emerged recently, such as\nquestion-answering and summarization. A common strategy to solve these tasks is\nto fine-tune... | 2403.09028 | 2404.00386 | 2403.09028_2404.00386 |
[
" Large language models (LLMs) have become pivotal in recent research. However,\nduring the inference process, LLMs still require substantial resources. In this\npaper, we propose CliqueParcel, a method designed to improve the efficiency of\nLLMs via prompt batching. Existing strategies to optimize inference effic... | 2402.14833 | 2404.07001 | 2402.14833_2404.07001 |
[
" In an era where artificial intelligence (AI) intertwines with medical\nresearch, the delineation of truth becomes increasingly complex. This study\nostensibly examines a purported novel SARS-CoV-2 variant, dubbed the Omega\nvariant, showcasing 31 unique mutations in the S gene region. However, the real\nundercur... | 2403.09674 | 2303.12665 | 2403.09674_2303.12665 |
[
" Large language models (LLMs) are effective at answering questions that are\nclearly asked. However, when faced with ambiguous queries they can act\nunpredictably and produce incorrect outputs. This underscores the need for the\ndevelopment of intelligent agents capable of asking clarification questions to\nresol... | 2310.01468 | 2211.00635 | 2310.01468_2211.00635 |
[
" Over the eight months since its release, ChatGPT and its underlying model,\nGPT3.5, have garnered massive attention, due to their potent mix of capability\nand accessibility. While a niche-industry of papers have emerged examining the\nscope of capabilities these models possess, the information fed to and\nextra... | 2307.16806 | 2402.08788 | 2307.16806_2402.08788 |
[
" This paper examines the challenges associated with achieving life-long\nsuperalignment in AI systems, particularly large language models (LLMs).\nSuperalignment is a theoretical framework that aspires to ensure that\nsuperintelligent AI systems act in accordance with human values and goals.\nDespite its promisin... | 2403.14683 | 2401.06466 | 2403.14683_2401.06466 |
[
" When interacting with Retrieval-Augmented Generation (RAG)-based\nconversational agents, the users must carefully craft their queries to be\nunderstood correctly. Yet, understanding the system's capabilities can be\nchallenging for the users, leading to ambiguous questions that necessitate\nfurther clarification... | 2403.11413 | 2404.02258 | 2403.11413_2404.02258 |
[
" This paper tackles recipe generation from unsegmented cooking videos, a task\nthat requires agents to (1) extract key events in completing the dish and (2)\ngenerate sentences for the extracted events. Our task is similar to dense video\ncaptioning (DVC), which aims at detecting events thoroughly and generating\... | 2209.10134 | 2402.15713 | 2209.10134_2402.15713 |
[
" Large language models (LLM) have achieved remarkable performance on various\nNLP tasks and are augmented by tools for broader applications. Yet, how to\nevaluate and analyze the tool-utilization capability of LLMs is still\nunder-explored. In contrast to previous works that evaluate models\nholistically, we comp... | 2312.14033 | 2401.03253 | 2312.14033_2401.03253 |
[
" Recently, decoder-only pre-trained large language models (LLMs), with several\ntens of billion parameters, have significantly impacted a wide range of natural\nlanguage processing (NLP) tasks. While encoder-only or encoder-decoder\npre-trained language models have already proved to be effective in discourse\npar... | 2403.05065 | 2306.17563 | 2403.05065_2306.17563 |
[
" In this work, we present HuixiangDou, a technical assistant powered by Large\nLanguage Models (LLM). This system is designed to assist algorithm developers\nby providing insightful responses to questions related to open-source algorithm\nprojects, such as computer vision and deep learning projects from OpenMMLab... | 2401.08772 | 2404.08259 | 2401.08772_2404.08259 |
[
" Generative Large Language Models (LLMs) have become the mainstream choice for\nfewshot and zeroshot learning thanks to the universality of text generation.\nMany users, however, do not need the broad capabilities of generative LLMs when\nthey only want to automate a classification task. Smaller BERT-like models ... | 2312.17543 | 2404.00913 | 2312.17543_2404.00913 |
[
" Electronic health records include information on patients' status and medical\nhistory, which could cover the history of diseases and disorders that could be\nhereditary. One important use of family history information is in precision\nhealth, where the goal is to keep the population healthy with preventative\nm... | 2403.09997 | 2403.00784 | 2403.09997_2403.00784 |
[
" This paper presents a question-answering approach to extract document-level\nevent-argument structures. We automatically ask and answer questions for each\nargument type an event may have. Questions are generated using manually defined\ntemplates and generative transformers. Template-based questions are generate... | 2404.16413 | 2307.10811 | 2404.16413_2307.10811 |
[
" The assessment of explainability in Legal Judgement Prediction (LJP) systems\nis of paramount importance in building trustworthy and transparent systems,\nparticularly considering the reliance of these systems on factors that may lack\nlegal relevance or involve sensitive attributes. This study delves into the\n... | 2402.17013 | 2402.10790 | 2402.17013_2402.10790 |
[
" The recent success in language generation capabilities of large language\nmodels (LLMs), such as GPT, Bard, Llama etc., can potentially lead to concerns\nabout their possible misuse in inducing mass agitation and communal hatred via\ngenerating fake news and spreading misinformation. Traditional means of\ndevelo... | 2401.04481 | 2403.01638 | 2401.04481_2403.01638 |
[
" End-to-end text spotting is a vital computer vision task that aims to\nintegrate scene text detection and recognition into a unified framework.\nTypical methods heavily rely on Region-of-Interest (RoI) operations to extract\nlocal features and complex post-processing steps to produce final predictions.\nTo addre... | 2306.03377 | 2403.04786 | 2306.03377_2403.04786 |
[
" In-context learning (ICL) is now a common method for teaching large language\nmodels (LLMs) new tasks: given labeled examples in the input context, the LLM\nlearns to perform the task without weight updates. Do models guided via ICL\ninfer the underlying structure of the task defined by the context, or do they\n... | 2311.07811 | 2401.15927 | 2311.07811_2401.15927 |
[
" This paper presents an overview of the PromptCBLUE shared task\n(http://cips-chip.org.cn/2023/eval1) held in the CHIP-2023 Conference. This\nshared task reformualtes the CBLUE benchmark, and provide a good testbed for\nChinese open-domain or medical-domain large language models (LLMs) in general\nmedical natural... | 2312.17522 | 2312.04127 | 2312.17522_2312.04127 |
[
" Knowledge Distillation (KD) is a predominant approach for BERT compression.\nPrevious KD-based methods focus on designing extra alignment losses for the\nstudent model to mimic the behavior of the teacher model. These methods\ntransfer the knowledge in an indirect way. In this paper, we propose a novel\nWeight-I... | 2305.09098 | 2403.12373 | 2305.09098_2403.12373 |
[
" The recent advances in deep-learning have led to the development of highly\nsophisticated systems with an unquenchable appetite for data. On the other\nhand, building good deep-learning models for low-resource languages remains a\nchallenging task. This paper focuses on developing a Question Answering dataset\nf... | 2308.09862 | 2312.14335 | 2308.09862_2312.14335 |
[
" Topic relevance between query and document is a very important part of social\nsearch, which can evaluate the degree of matching between document and user's\nrequirement. In most social search scenarios such as Dianping, modeling search\nrelevance always faces two challenges. One is that many documents in social... | 2404.02616 | 2404.00604 | 2404.02616_2404.00604 |
[
" In the last years' digitalization process, the creation and management of\ndocuments in various domains, particularly in Public Administration (PA), have\nbecome increasingly complex and diverse. This complexity arises from the need\nto handle a wide range of document types, often characterized by\nsemi-structur... | 2402.14871 | 2310.20620 | 2402.14871_2310.20620 |
[
" This paper investigates the inherent knowledge in language models from the\nperspective of epistemological holism. The purpose of this paper is to explore\nwhether LLMs exhibit characteristics consistent with epistemological holism.\nThese characteristics suggest that core knowledge, such as general scientific\n... | 2403.12862 | 2404.11449 | 2403.12862_2404.11449 |
[
" Recent years have seen important advances in the building of interpretable\nmodels, machine learning models that are designed to be easily understood by\nhumans. In this work, we show that large language models (LLMs) are remarkably\ngood at working with interpretable models, too. In particular, we show that\nLL... | 2402.14474 | 2403.17486 | 2402.14474_2403.17486 |
[
" We consider an unanswered question in the discourse processing community: why\ndo relation classifiers trained on explicit examples (with connectives removed)\nperform poorly in real implicit scenarios? Prior work claimed this is due to\nlinguistic dissimilarity between explicit and implicit examples but provide... | 2404.00999 | 2403.05065 | 2404.00999_2403.05065 |
[
" Human writers often bookend their writing with ending sentences that relate\nback to the beginning sentences in order to compose a satisfying narrative that\n\"closes the loop.\" Motivated by this observation, we propose RENarGen, a\ncontrollable story-generation paradigm that generates narratives by ensuring\nt... | 2404.00829 | 2210.05422 | 2404.00829_2210.05422 |
[
" Recent advances in named entity recognition (NER) have pushed the boundary of\nthe task to incorporate visual signals, leading to many variants, including\nmulti-modal NER (MNER) or grounded MNER (GMNER). A key challenge to these tasks\nis that the model should be able to generalize to the entities unseen during... | 2404.01914 | 2402.13036 | 2404.01914_2402.13036 |
[
" Adversarial purification is a defense mechanism for safeguarding classifiers\nagainst adversarial attacks without knowing the type of attacks or training of\nthe classifier. These techniques characterize and eliminate adversarial\nperturbations from the attacked inputs, aiming to restore purified samples that\nr... | 2402.06655 | 2308.06035 | 2402.06655_2308.06035 |
[
" Evaluating large language models (LLMs) is fundamental, particularly in the\ncontext of practical applications. Conventional evaluation methods, typically\ndesigned primarily for LLM development, yield numerical scores that ignore the\nuser experience. Therefore, our study shifts the focus from model-centered to... | 2404.07108 | 2310.12489 | 2404.07108_2310.12489 |
[
" With the proliferation of large pre-trained language models (PLMs),\nfine-tuning all model parameters becomes increasingly inefficient, particularly\nwhen dealing with numerous downstream tasks that entail substantial training\nand storage costs. Several approaches aimed at achieving parameter-efficient\nfine-tu... | 2401.11316 | 2301.00374 | 2401.11316_2301.00374 |
[
" How do language models \"think\"? This paper formulates a probabilistic\ncognitive model called the bounded pragmatic speaker, which can characterize\nthe operation of different variations of language models. Specifically, we\ndemonstrate that large language models fine-tuned with reinforcement learning\nfrom hu... | 2305.17760 | 2404.00950 | 2305.17760_2404.00950 |
[
" With the advent of large language models(LLMs) enhanced by the\nchain-of-thought(CoT) methodology, visual reasoning problem is usually\ndecomposed into manageable sub-tasks and tackled sequentially with various\nexternal tools. However, such a paradigm faces the challenge of the potential\n\"determining hallucin... | 2404.16033 | 2403.11838 | 2404.16033_2403.11838 |
[
" Large language models (LLMs) have gained significant attention in various\nfields but prone to hallucination, especially in knowledge-intensive (KI)\ntasks. To address this, retrieval-augmented generation (RAG) has emerged as a\npopular solution to enhance factual accuracy. However, traditional retrieval\nmodule... | 2402.01176 | 2309.16575 | 2402.01176_2309.16575 |
[
" The discourse around conspiracy theories is currently thriving amidst the\nrampant misinformation in online environments. Research in this field has been\nfocused on detecting conspiracy theories on social media, often relying on\nlimited datasets. In this study, we present a novel methodology for\nconstructing ... | 2308.15154 | 2311.10774 | 2308.15154_2311.10774 |
[
" Many models that leverage knowledge graphs (KGs) have recently demonstrated\nremarkable success in question answering (QA) tasks. In the real world, many\nfacts contained in KGs are time-constrained thus temporal KGQA has received\nincreasing attention. Despite the fruitful efforts of previous models in\ntempora... | 2402.13188 | 2402.14888 | 2402.13188_2402.14888 |
[
" Large Language Models (LLMs), such as ChatGPT, have drawn a lot of attentions\nrecently in the legal domain due to its emergent ability to tackle a variety of\nlegal tasks. However, it is still unknown if LLMs are able to analyze a legal\ncase and perform reasoning in the same manner as lawyers. Therefore, we\nc... | 2310.14880 | 2404.12491 | 2310.14880_2404.12491 |
[
" Recent advances in reinforcement learning (RL) algorithms aim to enhance the\nperformance of language models at scale. Yet, there is a noticeable absence of\na cost-effective and standardized testbed tailored to evaluating and comparing\nthese algorithms. To bridge this gap, we present a generalized version of t... | 2403.07191 | 2401.06105 | 2403.07191_2401.06105 |
[
" Efficiently fine-tuning Large Language Models (LLMs) for specific tasks\npresents a considerable challenge in natural language processing. Traditional\nmethods, like prompt or prefix tuning, typically rely on arbitrary tokens for\ntraining, leading to prolonged training times and generalized token use across\nva... | 2402.01643 | 2312.11803 | 2402.01643_2312.11803 |
[
" Large language models (LLMs) have achieved impressive performance in numerous\ndomains but often struggle to process lengthy inputs effectively and\nefficiently due to limited length generalization and attention's quadratic\ncomputational demands. Many sought to mitigate this by restricting the\nattention window... | 2402.10685 | 2404.18638 | 2402.10685_2404.18638 |
[
" Since humans can listen to audio and watch videos at faster speeds than\nactually observed, we often listen to or watch these pieces of content at\nhigher playback speeds to increase the time efficiency of content\ncomprehension. To further utilize this capability, systems that automatically\nadjust the playback... | 2403.02938 | 2401.13303 | 2403.02938_2401.13303 |
[
" Existing works focus on fixed-size layout pattern generation, while the more\npractical free-size pattern generation receives limited attention. In this\npaper, we propose ChatPattern, a novel Large-Language-Model (LLM) powered\nframework for flexible pattern customization. ChatPattern utilizes a two-part\nsyste... | 2403.15434 | 2402.01505 | 2403.15434_2402.01505 |
[
" Large Language Models (LLMs) have seen widespread deployment in various\nreal-world applications. Understanding these biases is crucial to comprehend\nthe potential downstream consequences when using LLMs to make decisions,\nparticularly for historically disadvantaged groups. In this work, we propose a\nsimple m... | 2308.02053 | 2307.00920 | 2308.02053_2307.00920 |
[
" The popularization of social media increases user engagements and generates a\nlarge amount of user-oriented data. Among them, text data (e.g., tweets, blogs)\nsignificantly attracts researchers and speculators to infer user attributes\n(e.g., age, gender, location) for fulfilling their intents. Generally, this\... | 2401.06802 | 2402.12080 | 2401.06802_2402.12080 |
[
" The task of identifying public opinions on social media, particularly\nregarding climate activism and the detection of hate events, has emerged as a\ncritical area of research in our rapidly changing world. With a growing number\nof people voicing either to support or oppose to climate-related issues -\nundersta... | 2402.01976 | 2404.18057 | 2402.01976_2404.18057 |
[
" Large Language Models (LLMs) can justify or critique their predictions\nthrough discussions with other models or humans, thereby enriching their\nintrinsic understanding of instances. While proactive discussions in the\ninference phase have been shown to boost performance, such interactions have\nnot been extens... | 2311.08107 | 2404.06137 | 2311.08107_2404.06137 |
[
" With the advent of larger and more complex deep learning models, such as in\nNatural Language Processing (NLP), model qualities like explainability and\ninterpretability, albeit highly desirable, are becoming harder challenges to\ntackle and solve. For example, state-of-the-art models in text classification\nare... | 2309.13340 | 2401.00689 | 2309.13340_2401.00689 |
[
" This document contains lectures and practical experimentations using Matlab\nand implementing a system which is actually correctly classifying three words\n(one, two and three) with the help of a very small database. To achieve this\nperformance, it uses speech modeling specificities, powerful computer\nalgorith... | 2402.01778 | 2309.13320 | 2402.01778_2309.13320 |
[
" The widespread use of pre-trained language models (PLMs) in natural language\nprocessing (NLP) has greatly improved performance outcomes. However, these\nmodels' vulnerability to adversarial attacks (e.g., camouflaged hints from drug\ndealers), particularly in the Chinese language with its rich character\ndivers... | 2404.12014 | 2311.04205 | 2404.12014_2311.04205 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.