Large-scale Language Models Research Articles

TC-BERT: large-scale language model for Korean technology commercialization documents

TC-BERT: large-scale language model for Korean technology commercialization documents

Advancements and challenges in AI-driven creative translation: A comprehensive analysis

Abstract. This article provides an in-depth exploration of the use of artificial intelligence (AI) in creative translation, highlighting the advancements in machine learning algorithms and the challenges of cultural and emotional congruence. It examines several key aspects: the efficacy of sequence-to-sequence and transformer models in enhancing translation accuracy, the role of AI in cultural contextualization, and the integration of emotional intelligence to handle subtleties in translations. Comparative case studies illustrate AI's performance against human translators in handling culturally dense texts, while adaptive algorithms demonstrate the customization of AI systems for specific cultural contexts. Furthermore, the article delves into the quantitative methods used to evaluate AI translation accuracy, such as BLEU and METEOR scores, and discusses innovative techniques for improving AI translation quality through reinforced learning and context-aware translations. The potential future directions of AI in creative translation, including the use of augmented reality and large-scale language models, are also discussed, presenting a vision of AI's evolving role in breaking linguistic and cultural barriers.

Integrating local knowledge with ChatGPT-like large-scale language models for enhanced societal comprehension of carbon neutrality

Addressing carbon neutrality presents a multifaceted challenge, necessitating collaboration across various disciplines, fields, and societal stakeholders. With the increasing urgency to mitigate climate change, there’s a crucial need for innovative approaches in communication and education to enhance societal understanding and engagement. Large-scale language models like ChatGPT emerge as transformative tools in the AI era, offering potential to revolutionize how we approach economic, technological, social, and environmental issues of achieving carbon neutrality. However, the full potential of these models in carbon neutrality is yet to be realized, hindered by limitations in providing detailed, localized, and expert-level insights across an expansive spectrum of subjects. To bridge these gaps, this paper introduces an innovative framework that integrates local knowledge with LLMs, aiming to markedly enhance the depth, accuracy, and regional relevance of the information provided. The effectiveness of this framework is examined from government, corporations, and community perspectives. The integration of local knowledge with LLMs not only enriches the AI’s comprehension of local specificities but also guarantees an up-to-date information that is crucial for addressing the specific concerns and questions about carbon neutrality raised by a broad array of stakeholders. Overall, the proposed framework showcases significant potential in enhancing societal comprehension and participation towards carbon neutrality.

Distributional hypothesis as isomorphism between word-word co-occurrence and analogical parallelograms.

Most of the modern natural language processing (NLP) techniques are based on the vector space models of language, in which each word is represented by a vector in a high dimensional space. One of the earliest successes was demonstrated by the four-term analogical reasoning task: what is to C as B is to A? The trained word vectors form "parallelograms" representing the quadruple of words in analogy. This discovery in NLP offers us insight into our understanding of human semantic representation of words via analogical reasoning. Despite successful applications of the large-scale language models, it has not been fully understood why such parallelograms emerge by learning through natural language data. As the vector space model is not optimized to form parallelograms, the key structure related to geometric shapes of word vectors is expected to be in the data, rather than the models. In the present article, we test our hypothesis that such parallelogram arrangement of word vectors readily exists in the co-occurrence statistics of language. Our approach focuses more on the data itself, and it is different from the existing theoretical approach trying to find the mechanism of parallelogram formation in the algorithms and/or vector arithmetic operations on word vectors. First, our analysis suggested that analogical reasoning is possible by decomposition of the bigram co-occurrence matrix. Second, we demonstrated the formation of a parallelepiped, a more structured geometric object than a parallelogram, by creating a small artificial corpus and its word vectors. With these results, we propose a refined form of distributional hypothesis pointing out an isomorphism between a sort of symmetry or exchangeability and word co-occurrence statistics.

MED-ChatGPT CoPilot: a ChatGPT medical assistant for case mining and adjunctive therapy.

The large-scale language model, GPT-4-1106-preview, supports text of up to 128 k characters, which has enhanced the capability of processing vast quantities of text. This model can perform efficient and accurate text data mining without the need for retraining, aided by prompt engineering. The research approach includes prompt engineering and text vectorization processing. In this study, prompt engineering is applied to assist ChatGPT in text mining. Subsequently, the mined results are vectorized and incorporated into a local knowledge base. After cleansing 306 medical papers, data extraction was performed using ChatGPT. Following a validation and filtering process, 241 medical case data entries were obtained, leading to the construction of a local medical knowledge base. Additionally, drawing upon the Langchain framework and utilizing the local knowledge base in conjunction with ChatGPT, we successfully developed a fast and reliable chatbot. This chatbot is capable of providing recommended diagnostic and treatment information for various diseases. The performance of the designed ChatGPT model, which was enhanced by data from the local knowledge base, exceeded that of the original model by 7.90% on a set of medical questions. ChatGPT, assisted by prompt engineering, demonstrates effective data mining capabilities for large-scale medical texts. In the future, we plan to incorporate a richer array of medical case data, expand the scale of the knowledge base, and enhance ChatGPT's performance in the medical field.

Semantic-aware frame-event fusion based pattern recognition via large vision–language models

Pattern recognition through the fusion of RGB frames and Event streams has emerged as a novel research area in recent years. Current methods typically employ backbone networks to individually extract the features of RGB frames and event streams, and subsequently fuse these features for pattern recognition. However, we posit that these methods may suffer from two key issues: (1). They attempt to directly learn a mapping from the input vision modality to the semantic labels. This approach often leads to sub-optimal results due to the disparity between the input and semantic labels; (2). They utilize small-scale backbone networks for the extraction of RGB and Event input features, thus these models fail to harness the recent performance advancements of large-scale visual-language models. In this study, we introduce a novel pattern recognition framework that consolidates the semantic labels, RGB frames, and event streams, leveraging pre-trained large-scale vision–language models. Specifically, given the input RGB frames, event streams, and all the predefined semantic labels, we employ a pre-trained large-scale vision model (CLIP vision encoder) to extract the RGB and event features. To handle the semantic labels, we initially convert them into language descriptions through prompt engineering and polish using ChatGPT, and then obtain the semantic features using the pre-trained large-scale language model (CLIP text encoder). Subsequently, we integrate the RGB/Event features and semantic features using multimodal Transformer networks. The resulting frame and event tokens are further amplified using self-attention layers. Concurrently, we propose to enhance the interactions between text tokens and RGB/Event tokens via cross-attention. Finally, we consolidate all three modalities using self-attention and feed-forward layers for recognition. Comprehensive experiments on the HARDVS and PokerEvent datasets fully substantiate the efficacy of our proposed SAFE model. The source code has been released at https://github.com/Event-AHU/SAFE_LargeVLM.

Military Equipment Entity Extraction Based on Large Language Model

The technology of military equipment entity extraction, a crucial component in constructing military knowledge bases, holds significant research value and theoretical importance for guiding the development and improvement of equipment support forces. In the military domain, equipment entities exhibit a phenomenon of nesting, where one entity is contained within another, and abbreviations or codes are frequently used to represent these entities. To address this complexity, this paper proposes a method named CoTNER for extracting entities. Initially, a large-scale language model is used to perform data augmentation with chain-of-thought on the original dataset, providing additional semantic and contextual information. Subsequently, the augmented dataset is fine-tuned on a small-scale language model to adapt it to the task of military equipment entity extraction and to enhance its ability to learn complex rules specific to the domain of military equipment. Additionally, a high-quality data filtering strategy based on instruction-following difficulty scoring is proposed to address the catastrophic forgetting issue that may occur during the fine-tuning of large language models. The experimental results demonstrate that the proposed military equipment entity extraction method outperforms mainstream traditional deep learning methods, validating the effectiveness of CoTNER.

Information Extraction of Aviation Accident Causation Knowledge Graph: An LLM-Based Approach

Summarizing the causation of aviation accidents is conducive to enhancing aviation safety. The knowledge graph of aviation accident causation, constructed based on aviation accident reports, can assist in analyzing the causes of aviation accidents. With the continuous development of artificial intelligence technology, leveraging large language models for information extraction and knowledge graph construction has demonstrated significant advantages. This paper proposes an information extraction method for aviation accident causation based on Claude-prompt, which relies on the large-scale pre-trained language model Claude 3.5. Through prompt engineering, combined with a few-shot learning strategy and a self-judgment mechanism, this method achieves automatic extraction of accident-cause entities and their relationships. Experimental results indicate that this approach effectively improves the accuracy of information extraction, overcoming the limitations of traditional methods in terms of accuracy and efficiency in processing complex texts. It provides strong support for subsequently constructing a structured knowledge graph of aviation accident causation and conducting causation analysis of aviation accidents.

Retrieval In Decoder benefits generative models for explainable complex question answering

Large-scale Language Models (LLMs) utilizing the Chain-of-Thought prompting demonstrate exceptional performance in a variety of tasks. However, the persistence of factual hallucinations remains a significant challenge in practical applications. Prevailing retrieval-augmented methods treat the retriever and generator as separate components, which inadvertently restricts the generator’s capabilities to those of the retriever through intensive supervised training. In this work, we propose an unsupervised Retrieval In Decoder framework for multi-granularity decoding called RID, which integrates retrieval directly into the decoding process of generative models. It dynamically adjusts decoding granularity based on retrieval outcomes, and duly corrects the decoding direction through its direct impact on the next token. Moreover, we introduce a reinforcement learning-driven knowledge distillation method for adaptive explanation generation to better apply to Small-scale Language Models (SLMs). The experimental results across six public benchmarks surpass popular LLMs and existing retrieval-augmented methods, which demonstrates the effectiveness of RID in models of different scales and verifies its applicability and scalability.

Utilizing ChatGPT for On-the-Fly Decision-Making in Autonomous Systems

In the realm of autonomous systems, rapid and efficient decision-making is critical to ensure optimal performance and adaptability in dynamic environments. This paper explores the application of ChatGPT, a large-scale language model, in on-the-fly decision-making within autonomous systems. By leveraging ChatGPT's capabilities in natural language processing, contextual understanding, and real-time response generation, the study demonstrates how it can support autonomous agents in interpreting situations, generating actionable insights, and making informed decisions. The integration of ChatGPT into autonomous systems provides a flexible and scalable approach to improving decision accuracy and response times in unpredictable scenarios. This research investigates the benefits, challenges, and potential use cases, highlighting the model's contributions to enhancing autonomy in various fields such as robotics, self-driving vehicles, and intelligent drones.

CMN: a co-designed neural architecture search for efficient computing-in-memory-based mixture-of-experts

Artificial intelligence (AI) has experienced substantial advancements recently, notably with the advent of large-scale language models (LLMs) employing mixture-of-experts (MoE) techniques, exhibiting human-like cognitive skills. As a promising hardware solution for edge MoE implementations, the computing-in-memory (CIM) architecture collocates memory and computing within a single device, significantly reducing the data movement and the associated energy consumption. However, due to diverse edge application scenarios and constraints, determining the optimal network structures for MoE, such as the expert’s location, quantity, and dimension on CIM systems remains elusive. To this end, we introduce a software-hardware co-designed neural architecture search (NAS) framework, CIM-based MoE NAS (CMN), focusing on identifying a high-performing MoE structure under specific hardware constraints. The results of the NYUD-v2 dataset segmentation on the RRAM (SRAM) CIM system reveal that CMN can discover optimized MoE configurations under energy, latency, and performance constraints, achieving 29.67× (43.10×) energy savings, 175.44×(109.89×) speedup, and 12.24× smaller model size compared to the baseline MoE-enabled Visual Transformer, respectively. This co-design opens up an avenue toward high-performance MoE deployments in edge CIM systems.

Prompt-supervised dynamic attention graph convolutional network for skeleton-based action recognition

Skeleton-based action recognition is a core task in the field of video understanding. Skeleton sequences are characterized by high information density, low redundancy, and clear structural information, thereby facilitating the analysis of complex relationships among human behaviors more readily than other modalities. Although existing studies have encoded skeleton data and achieved positive outcomes, they have often overlooked the precise high-level semantic information inherent in the action descriptions. To address this issue, this paper proposes a prompt-supervised dynamic attention graph convolutional network (PDA-GCN). Specifically, the PDA-GCN incorporates a prompt supervision (PS) module that leverages a pre-trained large-scale language model (LLM) as a knowledge engine and retains the generated text features as prompts to provide additional supervision during model training, enhancing the model’s ability to discern analogous actions with negligible computational cost. In addition, for the purpose of bolstering the learning of discriminative features, a dynamic attention graph convolution (DA-GC) module is presented. This module utilizes self-attention mechanism to adaptively infer intrinsic relationships between joints and integrates dynamic convolution to strengthen the emphasis on local information. This dual focus on both global context and local details further amplifies the efficiency and effectiveness of the model. Extensive experiments, conducted on the widely-used skeleton-based action recognition datasets NTU RGB+D 60 and NTU RGB+D 120, demonstrate that the PDA-GCN surpasses known state-of-the-art methods, achieving accuracies of 93.4% on the NTU RGB+D 60 cross-subject split and 90.7% on the NTU RGB+D 120 cross-subject split.

PDFll: Predictors of Disorder and Function of Proteins from the Language of Life.

The identification of intrinsically disordered proteins and their functional roles is largely dependent on the performance of computational predictors, necessitating a high standard of accuracy in these tools. In this context, we introduce a novel series of computational predictors, termed PDFll (Predictors of Disorder and Function of proteins from the Language of Life), which are designed to offer precise predictions of protein disorder and associated functional roles based on protein sequences. PDFll is developed through a two-step process. Initially, it leverages large-scale protein language models (pLMs), trained on an extensive dataset comprising billions of protein sequences. Subsequently, the embeddings derived from pLMs are integrated into streamlined, yet sophisticated, deep-learning models to generate predictions. These predictions notably surpass the performance of existing state-of-the-art predictors, particularly those that forecast disorder and function without utilizing evolutionary information.

Cross-lingual dependency parsing for a language with a unique script

Abstract Syntactic parsing is one of the areas in Natural Language Processing. The development of large-scale multilingual language models has enabled cross-lingual parsing approaches, which allows us to develop parsers for languages that do not have treebanks available. However, these approaches rely on the assumption that languages share orthographic representations and lexical entries. In this article, we investigate methods for developing a dependency parser for Xibe, a low-resource language that is written in a unique script. We first investigate lexicalized monolingual dependency parsing experiments to examine the effectiveness of word, part-of-speech, and character embeddings as well as pre-trained language models. Results show that character embeddings can significantly improve performance, while pre-trained language models decrease performance since they do not recognize the Xibe script. We also train delexicalized monolingual models, which yield competitive results to the best lexicalized model. Since the monolingual models are trained on a very small training set, we also investigate lexicalized and delexicalized cross-lingual models. We use six closely related languages as source language, which cover a wide range of scripts. In this setting, the delexicalized models achieve higher performance than lexicalized models. A final experiment shows that we can increase performance of the cross-lingual model by combining source languages and selecting the most similar sentences to Xibe as training set. However, all cross-lingual parsing results are still considerably lower than the monolingual model. We attribute the low performance of cross-lingual methods to syntactic and annotation differences as well as to the impoverished input of Universal Dependency Part-of-Speech tags that the delexicalized model has access to.

A Comprehensive Analysis of the per- and poly-fluoroalkyl substances (PFAS) research landscape through AI-assisted text mining

Per- and poly-fluoroalkyl substances (PFAS) have been widely used in various industrial applications due to their unique properties. This study aims to provide a comprehensive analysis of PFAS research trends using a novel approach combining text mining techniques and large-scale language models (LLMs). PFAS-related scientific literature published from 1980 to 2024 was gathered from Scopus, and KH Coder and Claude 3 were used to perform the analysis. The results showed a significant increase in research output and a clear shift in research topics over the past 40 years. Whereas in the past, the focus was on analytical methods, more recently, the emphasis has been on environmental fate, toxicity assessment, alternative compounds, and regulation. With Claude 3, research areas can now be identified without reviewing the results of expert text mining. Comparisons of AI-extracted trends with insights from traditional review articles showed strong agreement, confirming the effectiveness of this approach. These findings suggest the need for continued interdisciplinary research on PFAS such as the development of remediation strategies, elucidation of health effects, and evidence-based policymaking. This study showed the possibility of integrating text mining and LLM for a comprehensive analysis of research trends, which will accelerate future research and development strategies.

Preliminary assessment of TNM classification performance for pancreatic cancer in Japanese radiology reports using GPT-4.

A large-scale language model is expected to have been trained with a large volume of data including cancer treatment protocols. The current study aimed to investigate the use of generative pretrained transformer 4 (GPT-4) for identifying the TNM classification of pancreatic cancers from existing radiology reports written in Japanese. We screened 100 consecutive radiology reports on computed tomography scan for pancreatic cancer from April 2020 to June 2022. GPT-4 was requested to classify the TNM from the radiology reports based on the General Rules for the Study of Pancreatic Cancer 7th Edition. The accuracy and kappa coefficient of the TNM classifications by GPT-4 was evaluated with the classifications by two experienced abdominal radiologists as gold standard. The accuracy values of the T, N, and M factors were 0.73, 0.91, and 0.93, respectively. The kappa coefficients were 0.45 for T, 0.79 for N, and 0.83 for M. Although GPT is familiar with the TNM classification for pancreatic cancer, its performance in classifying actual cases in this experiment may not be adequate.

The impact of fine-tuning paradigms on unknown plant diseases recognition

Plant diseases pose significant threats to agriculture, impacting both food safety and public health. Traditional plant disease detection systems are typically limited to recognizing disease categories included in the training dataset, rendering them ineffective against new disease types. Although out-of-distribution (OOD) detection methods have been proposed to address this issue, the impact of fine-tuning paradigms on these methods has been overlooked. This paper focuses on studying the impact of fine-tuning paradigms on the performance of detecting unknown plant diseases. Currently, fine-tuning on visual tasks is mainly divided into visual-based models and visual-language-based models. We first discuss the limitations of large-scale visual language models in this task: textual prompts are difficult to design. To avoid the side effects of textual prompts, we futher explore the effectiveness of purely visual pre-trained models for OOD detection in plant disease tasks. Specifically, we employed five publicly accessible datasets to establish benchmarks for open-set recognition, OOD detection, and few-shot learning in plant disease recognition. Additionally, we comprehensively compared various OOD detection methods, fine-tuning paradigms, and factors affecting OOD detection performance, such as sample quantity. The results show that visual prompt tuning outperforms fully fine-tuning and linear probe tuning in out-of-distribution detection performance, especially in the few-shot scenarios. Notably, the max-logit-based on visual prompt tuning achieves an AUROC score of 94.8%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document} in the 8-shot setting, which is nearly comparable to the method of fully fine-tuning on the full dataset (95.2%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document}), which implies that an appropriate fine-tuning paradigm can directly improve OOD detection performance. Finally, we visualized the prediction distributions of different OOD detection methods and discussed the selection of thresholds. Overall, this work lays the foundation for unknown plant disease recognition, providing strong support for the security and reliability of plant disease recognition systems. We will release our code at https://github.com/JiuqingDong/PDOOD to further advance this field.

A Human Feedback Strategy for Photoresponsive Molecules in Drug Delivery: Utilizing GPT-2 and Time-Dependent Density Functional Theory Calculations.

Photoresponsive drug delivery stands as a pivotal frontier in smart drug administration, leveraging the non-invasive, stable, and finely tunable nature of light-triggered methodologies. The generative pre-trained transformer (GPT) has been employed to generate molecular structures. In our study, we harnessed GPT-2 on the QM7b dataset to refine a UV-GPT model with adapters, enabling the generation of molecules responsive to UV light excitation. Utilizing the Coulomb matrix as a molecular descriptor, we predicted the excitation wavelengths of these molecules. Furthermore, we validated the excited state properties through quantum chemical simulations. Based on the results of these calculations, we summarized some tips for chemical structures and integrated them into the alignment of large-scale language models within the reinforcement learning from human feedback (RLHF) framework. The synergy of these findings underscores the successful application of GPT technology in this critical domain.

Emergency Patient Triage Improvement through a Retrieval-Augmented Generation Enhanced Large-Scale Language Model

Objectives Emergency medical triage is crucial for prioritizing patient care in emergency situations, yet its effectiveness can vary significantly based on the experience and training of the personnel involved. This study aims to evaluate the efficacy of integrating Retrieval Augmented Generation (RAG) with Large Language Models (LLMs), specifically OpenAI's GPT models, to standardize triage procedures and reduce variability in emergency care. Methods We created 100 simulated triage scenarios based on modified cases from the Japanese National Examination for Emergency Medical Technicians. These scenarios were processed by the RAG-enhanced LLMs, and the models were given patient vital signs, symptoms, and observations from emergency medical services (EMS) teams as inputs. The primary outcome was the accuracy of triage classifications, which was used to compare the performance of the RAG-enhanced LLMs with that of emergency medical technicians and emergency physicians. Secondary outcomes included the rates of under-triage and over-triage. Results The Generative Pre-trained Transformer 3.5 (GPT-3.5) with RAG model achieved a correct triage rate of 70%, significantly outperforming Emergency Medical Technicians (EMTs) with 35% and 38% correct rates, and emergency physicians with 50% and 47% correct rates (p < 0.05). Additionally, this model demonstrated a substantial reduction in under-triage rates to 8%, compared with 33% for GPT-3.5 without RAG, and 39% for GPT-4 without RAG. Conclusions The integration of RAG with LLMs shows promise in improving the accuracy and consistency of medical assessments in emergency settings. Further validation in diverse medical settings with broader datasets is necessary to confirm the effectiveness and adaptability of these technologies in live environments.

Assessing Accuracy of ChatGPT on Addressing Helicobacter pylori Infection-Related Questions: A National Survey and Comparative Study.

ChatGPT is a novel and online large-scale language model used as a source providing up-to-date and useful health-related knowledges to patients and clinicians. However, its performance on Helicobacter pylori infection-related questions remain unknown. This study aimed to evaluate the accuracy of ChatGPT's responses on H. pylori-related questions compared with that of gastroenterologists during the same period. Twenty-five H. pylori-related questions from five domains: Indication, Diagnostics, Treatment, Gastric cancer and prevention, and Gut Microbiota were selected based on the Maastricht VI Consensus report. Each question was tested three times with ChatGPT3.5 and ChatGPT4. Two independent H. pylori experts assessed the responses from ChatGPT, with discrepancies resolved by a third reviewer. Simultaneously, a nationwide survey with the same questions was conducted among 1279 gastroenterologists and 154 medical students. The accuracy of responses from ChatGPT3.5 and ChatGPT4 was compared with that of gastroenterologists. Overall, both ChatGPT3.5 and ChatGPT4 demonstrated high accuracy, with median accuracy rates of 92% for each of the three responses, surpassing the accuracy of nationwide gastroenterologists (median: 80%) and equivalent to that of senior gastroenterologists. Compared with ChatGPT3.5, ChatGPT4 provided more concise responses with the same accuracy. ChatGPT3.5 performed well in the Indication, Treatment, and Gut Microbiota domains, whereas ChatGPT4 excelled in Diagnostics, Gastric cancer and prevention, and Gut Microbiota domains. ChatGPT exhibited high accuracy and reproducibility in addressing H. pylori-related questions except the decision for H. pylori treatment, performing at the level of senior gastroenterologists and could serve as an auxiliary information tool for assisting patients and clinicians.