The excavation of deep tunnels crossing faults is highly prone to triggering rockburst disasters,which has become a significant engineering issue.In this study,taking the fault-slip rockbursts from a deep tunnel in so...The excavation of deep tunnels crossing faults is highly prone to triggering rockburst disasters,which has become a significant engineering issue.In this study,taking the fault-slip rockbursts from a deep tunnel in southwestern China as the engineering prototype,large-scale three-dimensional(3D)physical model tests were conducted on a 3D-printed complex geological model containing two faults.Based on the selfdeveloped 3D loading system and excavation device,the macroscopic failure of fault-slip rockbursts was simulated indoors.The stress,strain,and fracturing characteristics of the surrounding rock near the two faults were systematically evaluated during excavation and multistage loading.The test results effectively revealed the evolution and triggering mechanism of fault-slip rockbursts.After the excavation of a highstress tunnel,stress readjustment occurred.Owing to the presence of these two faults,stress continued to accumulate in the rock mass between them,leading to the accumulation of fractures.When the shear stress on a fault surface exceeded its shear strength,sudden fault slip and dislocation occurred,thus triggering rockbursts.Rockbursts occurred twice in the vault between the two faults,showing obvious intermittent characteristics.The rockburst pit was controlled by two faults.When the faults remained stable,tensile failure predominated in the surrounding rock.However,when the fault slip was triggered,shear failure in the surrounding rock increased.These findings provide valuable insights for enhancing the comprehension of fault-slip rockbursts.展开更多
The aim of this article is to explore potential directions for the development of artificial intelligence(AI).It points out that,while current AI can handle the statistical properties of complex systems,it has difficu...The aim of this article is to explore potential directions for the development of artificial intelligence(AI).It points out that,while current AI can handle the statistical properties of complex systems,it has difficulty effectively processing and fully representing their spatiotemporal complexity patterns.The article also discusses a potential path of AI development in the engineering domain.Based on the existing understanding of the principles of multilevel com-plexity,this article suggests that consistency among the logical structures of datasets,AI models,model-building software,and hardware will be an important AI development direction and is worthy of careful consideration.展开更多
This paper proposes a non-intrusive computational method for mechanical dynamic systems involving a large-scale of interval uncertain parameters,aiming to reduce the computational costs and improve accuracy in determi...This paper proposes a non-intrusive computational method for mechanical dynamic systems involving a large-scale of interval uncertain parameters,aiming to reduce the computational costs and improve accuracy in determining bounds of system response.The screening method is firstly used to reduce the scale of active uncertain parameters.The sequential high-order polynomials surrogate models are then used to approximate the dynamic system’s response at each time step.To reduce the sampling cost of constructing surrogate model,the interaction effect among uncertain parameters is gradually added to the surrogate model by sequentially incorporating samples from a candidate set,which is composed of vertices and inner grid points.Finally,the points that may produce the bounds of the system response at each time step are searched using the surrogate models.The optimization algorithm is used to locate extreme points,which contribute to determining the inner points producing system response bounds.Additionally,all vertices are also checked using the surrogate models.A vehicle nonlinear dynamic model with 72 uncertain parameters is presented to demonstrate the accuracy and efficiency of the proposed uncertain computational method.展开更多
With the rapid development of large AI models,large decision models have further broken through the limits of human cognition and promoted the innovation of decision-making paradigms in extensive fields such as medici...With the rapid development of large AI models,large decision models have further broken through the limits of human cognition and promoted the innovation of decision-making paradigms in extensive fields such as medicine and transportation.In this paper,we systematically expound on the intelligent decision-making technology and prospects driven by large AI models.Specifically,we first review the development of large AI models in recent years.Then,from the perspective of methods,we introduce important theories and technologies of large decision models,such as model architecture and model adaptation.Next,from the perspective of applications,we introduce the cutting-edge applications of large decision models in various fields,such as autonomous driving and knowledge decision-making.Finally,we discuss existing challenges,such as security issues,decision bias and hallucination phenomenon as well as future prospects,from both technology development and domain applications.We hope this review paper can help researchers understand the important progress of intelligent decision-making driven by large AI models.展开更多
A Long Short-Term Memory(LSTM) Recurrent Neural Network(RNN) has driven tremendous improvements on an acoustic model based on Gaussian Mixture Model(GMM). However, these models based on a hybrid method require a force...A Long Short-Term Memory(LSTM) Recurrent Neural Network(RNN) has driven tremendous improvements on an acoustic model based on Gaussian Mixture Model(GMM). However, these models based on a hybrid method require a forced aligned Hidden Markov Model(HMM) state sequence obtained from the GMM-based acoustic model. Therefore, it requires a long computation time for training both the GMM-based acoustic model and a deep learning-based acoustic model. In order to solve this problem, an acoustic model using CTC algorithm is proposed. CTC algorithm does not require the GMM-based acoustic model because it does not use the forced aligned HMM state sequence. However, previous works on a LSTM RNN-based acoustic model using CTC used a small-scale training corpus. In this paper, the LSTM RNN-based acoustic model using CTC is trained on a large-scale training corpus and its performance is evaluated. The implemented acoustic model has a performance of 6.18% and 15.01% in terms of Word Error Rate(WER) for clean speech and noisy speech, respectively. This is similar to a performance of the acoustic model based on the hybrid method.展开更多
DeepSeek,a Chinese artificial intelligence(AI)startup,has released their V3 and R1 series models,which attracted global attention due to their low cost,high performance,and open-source advantages.This paper begins by ...DeepSeek,a Chinese artificial intelligence(AI)startup,has released their V3 and R1 series models,which attracted global attention due to their low cost,high performance,and open-source advantages.This paper begins by reviewing the evolution of large AI models focusing on paradigm shifts,the mainstream large language model(LLM)paradigm,and the DeepSeek paradigm.Subsequently,the paper highlights novel algorithms introduced by DeepSeek,including multi-head latent attention(MLA),mixture-of-experts(MoE),multi-token prediction(MTP),and group relative policy optimization(GRPO).The paper then explores DeepSeek's engineering breakthroughs in LLM scaling,training,inference,and system-level optimization architecture.Moreover,the impact of DeepSeek models on the competitive AI landscape is analyzed,comparing them to mainstream LLMs across various fields.Finally,the paper reflects on the insights gained from DeepSeek's innovations and discusses future trends in the technical and engineering development of large AI models,particularly in data,training,and reasoning.展开更多
In the wave of the“Internet+AI”era,information technology is comprehensively reshaping the landscape of college English reading education.Traditional teaching models struggle to meet the demands of talent cultivatio...In the wave of the“Internet+AI”era,information technology is comprehensively reshaping the landscape of college English reading education.Traditional teaching models struggle to meet the demands of talent cultivation in the new era.The integration of“Internet+AI”technologies brings revolutionary opportunities to reading instruction,significantly enriching teaching resources,enabling personalized teaching,enhancing interactivity,and optimizing evaluation systems.Guided by principles such as student-centeredness and integrated innovation,this study proposes multiple strategies for advancing teaching practices.Using the Understanding Contemporary China:English Reading and Writing Tutorial(Foreign Language Teaching and Research Press)as a case study,this paper explores practical pathways for reforming college English reading instruction,aiming to improve teaching quality and students’comprehensive English reading literacy.展开更多
The results of mass appraisal in many countries are used as a basis for calculating the amount of real estate tax,therefore,regardless of the methods used to calculate it,the resulting value should be as close as poss...The results of mass appraisal in many countries are used as a basis for calculating the amount of real estate tax,therefore,regardless of the methods used to calculate it,the resulting value should be as close as possible to the market value of the real estate to maintain a balance of interests between the state and the rights holders.In practice,this condition is not always met,since,firstly,the quality of market data is often very low,and secondly,some markets are characterized by low activity,which is expressed in a deficit of information on asking prices.The aim of the work is ecological valuation of land use:how regression-based mass appraisal can inform ecological conservation,land degradation,and sustainable land management.Four multiple regression models were constructed for AI generated map of land plots for recreational use in St.Petersburg(Russia)with different volumes of market information(32,30,20 and 15 units of market information with four price-forming factors).During the analysis of the quality of the models,it was revealed that the best result is shown by the model built on the maximum sample size,then the model based on 15 analogs,which proves that a larger number of analog objects does not always allow us to achieve better results,since the more analog objects there are.展开更多
Large language models(LLMs)have undergone significant expansion and have been increasingly integrated across various domains.Notably,in the realm of robot task planning,LLMs harness their advanced reasoning and langua...Large language models(LLMs)have undergone significant expansion and have been increasingly integrated across various domains.Notably,in the realm of robot task planning,LLMs harness their advanced reasoning and language comprehension capabilities to formulate precise and efficient action plans based on natural language instructions.However,for embodied tasks,where robots interact with complex environments,textonly LLMs often face challenges due to a lack of compatibility with robotic visual perception.This study provides a comprehensive overview of the emerging integration of LLMs and multimodal LLMs into various robotic tasks.Additionally,we propose a framework that utilizes multimodal GPT-4V to enhance embodied task planning through the combination of natural language instructions and robot visual perceptions.Our results,based on diverse datasets,indicate that GPT-4V effectively enhances robot performance in embodied tasks.This extensive survey and evaluation of LLMs and multimodal LLMs across a variety of robotic tasks enriches the understanding of LLM-centric embodied intelligence and provides forward-looking insights towards bridging the gap in Human-Robot-Environment interaction.展开更多
The streamflow over the Yellow River basin is simulated using the PRECIS （Providing REgional Climates for Impacts Studies） regional climate model driven by 15-year （1979-1993） ECMWF reanalysis data as the initial ...The streamflow over the Yellow River basin is simulated using the PRECIS （Providing REgional Climates for Impacts Studies） regional climate model driven by 15-year （1979-1993） ECMWF reanalysis data as the initial and lateral boundary conditions and an off-line large-scale routing model （LRM）. The LRM uses physical catchment and river channel information and allows streamflow to be predicted for large continental rivers with a 1°×1° spatial resolution. The results show that the PRECIS model can reproduce the general southeast to northwest gradient distribution of the precipitation over the Yellow River basin, The PRECIS- LRM model combination has the capability to simulate the seasonal and annual streamflow over the Yellow River basin. The simulated streamflow is generally coincident with the naturalized streamflow both in timing and in magnitude.展开更多
ChatGPT is a powerful artificial intelligence(AI)language model that has demonstrated significant improvements in various natural language processing(NLP) tasks. However, like any technology, it presents potential sec...ChatGPT is a powerful artificial intelligence(AI)language model that has demonstrated significant improvements in various natural language processing(NLP) tasks. However, like any technology, it presents potential security risks that need to be carefully evaluated and addressed. In this survey, we provide an overview of the current state of research on security of using ChatGPT, with aspects of bias, disinformation, ethics, misuse,attacks and privacy. We review and discuss the literature on these topics and highlight open research questions and future directions.Through this survey, we aim to contribute to the academic discourse on AI security, enriching the understanding of potential risks and mitigations. We anticipate that this survey will be valuable for various stakeholders involved in AI development and usage, including AI researchers, developers, policy makers, and end-users.展开更多
Large-scale Language Models(LLMs)have achieved significant breakthroughs in Natural Language Processing(NLP),driven by the pre-training and fine-tuning paradigm.While this approach allows models to specialize in speci...Large-scale Language Models(LLMs)have achieved significant breakthroughs in Natural Language Processing(NLP),driven by the pre-training and fine-tuning paradigm.While this approach allows models to specialize in specific tasks with reduced training costs,the substantial memory requirements during fine-tuning present a barrier to broader deployment.Parameter-Efficient Fine-Tuning(PEFT)techniques,such as Low-Rank Adaptation(LoRA),and parameter quantization methods have emerged as solutions to address these challenges by optimizing memory usage and computational efficiency.Among these,QLoRA,which combines PEFT and quantization,has demonstrated notable success in reducing memory footprints during fine-tuning,prompting the development of various QLoRA variants.Despite these advancements,the quantitative impact of key variables on the fine-tuning performance of quantized LLMs remains underexplored.This study presents a comprehensive analysis of these key variables,focusing on their influence across different layer types and depths within LLM architectures.Our investigation uncovers several critical findings:(1)Larger layers,such as MLP layers,can maintain performance despite reductions in adapter rank,while smaller layers,like self-attention layers,aremore sensitive to such changes;(2)The effectiveness of balancing factors depends more on specific values rather than layer type or depth;(3)In quantization-aware fine-tuning,larger layers can effectively utilize smaller adapters,whereas smaller layers struggle to do so.These insights suggest that layer type is a more significant determinant of fine-tuning success than layer depth when optimizing quantized LLMs.Moreover,for the same discount of trainable parameters,reducing the trainable parameters in a larger layer is more effective in preserving fine-tuning accuracy than in a smaller one.This study provides valuable guidance for more efficient fine-tuning strategies and opens avenues for further research into optimizing LLM fine-tuning in resource-constrained environments.展开更多
Oral expression skills play an essential role in the development of EFL students’language abilities,and how to improve EFL students’oral expression skills is an essential and challenging task.This study adopts a qua...Oral expression skills play an essential role in the development of EFL students’language abilities,and how to improve EFL students’oral expression skills is an essential and challenging task.This study adopts a quasi-experimental research method to carry out the research and proposes an AI-based reflective dialogue model.Based on this,an analysis of the impact brought by this model on EFL students’oral expression performance and learning anxiety levels.The results show that students in the experimental group have significantly higher oral expression performance than those in the control group in the three dimensions of grammatical accuracy,expressive fluency,and word accuracy.In addition,the students in the experimental group produced facilitated anxiety after using the AI-based reflective dialogue model for oral expression learning,which prompted the students to learn more diligently.展开更多
Bedding slope is a typical heterogeneous slope consisting of different soil/rock layers and is likely to slide along the weakest interface.Conventional slope protection methods for bedding slopes,such as retaining wal...Bedding slope is a typical heterogeneous slope consisting of different soil/rock layers and is likely to slide along the weakest interface.Conventional slope protection methods for bedding slopes,such as retaining walls,stabilizing piles,and anchors,are time-consuming and labor-and energy-intensive.This study proposes an innovative polymer grout method to improve the bearing capacity and reduce the displacement of bedding slopes.A series of large-scale model tests were carried out to verify the effectiveness of polymer grout in protecting bedding slopes.Specifically,load-displacement relationships and failure patterns were analyzed for different testing slopes with various dosages of polymer.Results show the great potential of polymer grout in improving bearing capacity,reducing settlement,and protecting slopes from being crushed under shearing.The polymer-treated slopes remained structurally intact,while the untreated slope exhibited considerable damage when subjected to loads surpassing the bearing capacity.It is also found that polymer-cemented soils concentrate around the injection pipe,forming a fan-shaped sheet-like structure.This study proves the improvement of polymer grouting for bedding slope treatment and will contribute to the development of a fast method to protect bedding slopes from landslides.展开更多
Considering the large diameter effect of piles,the influence of different pile-soil analysis methods on the design of monopile foundations for offshore wind turbines has become an urgent problem to be solved.Three dif...Considering the large diameter effect of piles,the influence of different pile-soil analysis methods on the design of monopile foundations for offshore wind turbines has become an urgent problem to be solved.Three different pile-soil models were used to study a large 10 MW monopile wind turbine.By modeling the three models in the SACS software,this paper analyzed the motion response of the overall structure under the conditions of wind and waves.According to the given working conditions,this paper concludes that under the condition of independent wind,the average value of the tower top x-displacement of the rigid connection method is the smalle st,and the standard deviation is the smallest under the condition of independent wave.The results obtained by the p-y curve method are the most conservative.展开更多
This paper investigates the wireless communication with a novel architecture of antenna arrays,termed modular extremely large-scale array(XLarray),where array elements of an extremely large number/size are regularly m...This paper investigates the wireless communication with a novel architecture of antenna arrays,termed modular extremely large-scale array(XLarray),where array elements of an extremely large number/size are regularly mounted on a shared platform with both horizontally and vertically interlaced modules.Each module consists of a moderate/flexible number of array elements with the inter-element distance typically in the order of the signal wavelength,while different modules are separated by the relatively large inter-module distance for convenience of practical deployment.By accurately modelling the signal amplitudes and phases,as well as projected apertures across all modular elements,we analyse the near-field signal-to-noise ratio(SNR)performance for modular XL-array communications.Based on the non-uniform spherical wave(NUSW)modelling,the closed-form SNR expression is derived in terms of key system parameters,such as the overall modular array size,distances of adjacent modules along all dimensions,and the user's three-dimensional(3D)location.In addition,with the number of modules in different dimensions increasing infinitely,the asymptotic SNR scaling laws are revealed.Furthermore,we show that our proposed near-field modelling and performance analysis include the results for existing array architectures/modelling as special cases,e.g.,the collocated XL-array architecture,the uniform plane wave(UPW)based far-field modelling,and the modular extremely large-scale uniform linear array(XL-ULA)of onedimension.Extensive simulation results are presented to validate our findings.展开更多
基金funding support from the National Natural Science Foundation of China(Grant Nos.42177136 and 52309126).
文摘The excavation of deep tunnels crossing faults is highly prone to triggering rockburst disasters,which has become a significant engineering issue.In this study,taking the fault-slip rockbursts from a deep tunnel in southwestern China as the engineering prototype,large-scale three-dimensional(3D)physical model tests were conducted on a 3D-printed complex geological model containing two faults.Based on the selfdeveloped 3D loading system and excavation device,the macroscopic failure of fault-slip rockbursts was simulated indoors.The stress,strain,and fracturing characteristics of the surrounding rock near the two faults were systematically evaluated during excavation and multistage loading.The test results effectively revealed the evolution and triggering mechanism of fault-slip rockbursts.After the excavation of a highstress tunnel,stress readjustment occurred.Owing to the presence of these two faults,stress continued to accumulate in the rock mass between them,leading to the accumulation of fractures.When the shear stress on a fault surface exceeded its shear strength,sudden fault slip and dislocation occurred,thus triggering rockbursts.Rockbursts occurred twice in the vault between the two faults,showing obvious intermittent characteristics.The rockburst pit was controlled by two faults.When the faults remained stable,tensile failure predominated in the surrounding rock.However,when the fault slip was triggered,shear failure in the surrounding rock increased.These findings provide valuable insights for enhancing the comprehension of fault-slip rockbursts.
文摘The aim of this article is to explore potential directions for the development of artificial intelligence(AI).It points out that,while current AI can handle the statistical properties of complex systems,it has difficulty effectively processing and fully representing their spatiotemporal complexity patterns.The article also discusses a potential path of AI development in the engineering domain.Based on the existing understanding of the principles of multilevel com-plexity,this article suggests that consistency among the logical structures of datasets,AI models,model-building software,and hardware will be an important AI development direction and is worthy of careful consideration.
基金supported by the National Natural Science Foundation of China(Grant No.12272142)Fundamental Research Funds for the Central Universities(Grant No.2172021XXJS048)。
文摘This paper proposes a non-intrusive computational method for mechanical dynamic systems involving a large-scale of interval uncertain parameters,aiming to reduce the computational costs and improve accuracy in determining bounds of system response.The screening method is firstly used to reduce the scale of active uncertain parameters.The sequential high-order polynomials surrogate models are then used to approximate the dynamic system’s response at each time step.To reduce the sampling cost of constructing surrogate model,the interaction effect among uncertain parameters is gradually added to the surrogate model by sequentially incorporating samples from a candidate set,which is composed of vertices and inner grid points.Finally,the points that may produce the bounds of the system response at each time step are searched using the surrogate models.The optimization algorithm is used to locate extreme points,which contribute to determining the inner points producing system response bounds.Additionally,all vertices are also checked using the surrogate models.A vehicle nonlinear dynamic model with 72 uncertain parameters is presented to demonstrate the accuracy and efficiency of the proposed uncertain computational method.
基金supported by the National Natural Science Foundation of China(Grant 62293545)Shenzhen Science and Technology Program(Grant ZDSYS20220323112000001).
文摘With the rapid development of large AI models,large decision models have further broken through the limits of human cognition and promoted the innovation of decision-making paradigms in extensive fields such as medicine and transportation.In this paper,we systematically expound on the intelligent decision-making technology and prospects driven by large AI models.Specifically,we first review the development of large AI models in recent years.Then,from the perspective of methods,we introduce important theories and technologies of large decision models,such as model architecture and model adaptation.Next,from the perspective of applications,we introduce the cutting-edge applications of large decision models in various fields,such as autonomous driving and knowledge decision-making.Finally,we discuss existing challenges,such as security issues,decision bias and hallucination phenomenon as well as future prospects,from both technology development and domain applications.We hope this review paper can help researchers understand the important progress of intelligent decision-making driven by large AI models.
基金supported by the Ministry of Trade,Industry & Energy(MOTIE,Korea) under Industrial Technology Innovation Program (No.10063424,'development of distant speech recognition and multi-task dialog processing technologies for in-door conversational robots')
文摘A Long Short-Term Memory(LSTM) Recurrent Neural Network(RNN) has driven tremendous improvements on an acoustic model based on Gaussian Mixture Model(GMM). However, these models based on a hybrid method require a forced aligned Hidden Markov Model(HMM) state sequence obtained from the GMM-based acoustic model. Therefore, it requires a long computation time for training both the GMM-based acoustic model and a deep learning-based acoustic model. In order to solve this problem, an acoustic model using CTC algorithm is proposed. CTC algorithm does not require the GMM-based acoustic model because it does not use the forced aligned HMM state sequence. However, previous works on a LSTM RNN-based acoustic model using CTC used a small-scale training corpus. In this paper, the LSTM RNN-based acoustic model using CTC is trained on a large-scale training corpus and its performance is evaluated. The implemented acoustic model has a performance of 6.18% and 15.01% in terms of Word Error Rate(WER) for clean speech and noisy speech, respectively. This is similar to a performance of the acoustic model based on the hybrid method.
基金supported by the National Natural Science Foundation of China(62233005,62293502,U2441245,62176185,U23B2057,62306112)the STCSM Science and Technology Innovation Action Plan Computational Biology Program(24JS2830400)+2 种基金the State Key Laboratory of Industrial Control Technology,China(ICT2024A22)the Shanghai Sailing Program(23YF1409400)the National Science and Technology Major Project(2024ZD0532403).
文摘DeepSeek,a Chinese artificial intelligence(AI)startup,has released their V3 and R1 series models,which attracted global attention due to their low cost,high performance,and open-source advantages.This paper begins by reviewing the evolution of large AI models focusing on paradigm shifts,the mainstream large language model(LLM)paradigm,and the DeepSeek paradigm.Subsequently,the paper highlights novel algorithms introduced by DeepSeek,including multi-head latent attention(MLA),mixture-of-experts(MoE),multi-token prediction(MTP),and group relative policy optimization(GRPO).The paper then explores DeepSeek's engineering breakthroughs in LLM scaling,training,inference,and system-level optimization architecture.Moreover,the impact of DeepSeek models on the competitive AI landscape is analyzed,comparing them to mainstream LLMs across various fields.Finally,the paper reflects on the insights gained from DeepSeek's innovations and discusses future trends in the technical and engineering development of large AI models,particularly in data,training,and reasoning.
文摘In the wave of the“Internet+AI”era,information technology is comprehensively reshaping the landscape of college English reading education.Traditional teaching models struggle to meet the demands of talent cultivation in the new era.The integration of“Internet+AI”technologies brings revolutionary opportunities to reading instruction,significantly enriching teaching resources,enabling personalized teaching,enhancing interactivity,and optimizing evaluation systems.Guided by principles such as student-centeredness and integrated innovation,this study proposes multiple strategies for advancing teaching practices.Using the Understanding Contemporary China:English Reading and Writing Tutorial(Foreign Language Teaching and Research Press)as a case study,this paper explores practical pathways for reforming college English reading instruction,aiming to improve teaching quality and students’comprehensive English reading literacy.
基金financed as part of the project“Development of a methodology for instrumental base formation for analysis and modeling of the spatial socio-economic development of systems based on internal reserves in the context of digitalization”(FSEG-2023-0008)funded by the Russian Science Foundation(Agreement 23-41-10001,https://doi.org/https://rscf.ru/project/23-41-10001/).
文摘The results of mass appraisal in many countries are used as a basis for calculating the amount of real estate tax,therefore,regardless of the methods used to calculate it,the resulting value should be as close as possible to the market value of the real estate to maintain a balance of interests between the state and the rights holders.In practice,this condition is not always met,since,firstly,the quality of market data is often very low,and secondly,some markets are characterized by low activity,which is expressed in a deficit of information on asking prices.The aim of the work is ecological valuation of land use:how regression-based mass appraisal can inform ecological conservation,land degradation,and sustainable land management.Four multiple regression models were constructed for AI generated map of land plots for recreational use in St.Petersburg(Russia)with different volumes of market information(32,30,20 and 15 units of market information with four price-forming factors).During the analysis of the quality of the models,it was revealed that the best result is shown by the model built on the maximum sample size,then the model based on 15 analogs,which proves that a larger number of analog objects does not always allow us to achieve better results,since the more analog objects there are.
基金supported by National Natural Science Foundation of China(62376219 and 62006194)Foundational Research Project in Specialized Discipline(Grant No.G2024WD0146)Faculty Construction Project(Grant No.24GH0201148).
文摘Large language models(LLMs)have undergone significant expansion and have been increasingly integrated across various domains.Notably,in the realm of robot task planning,LLMs harness their advanced reasoning and language comprehension capabilities to formulate precise and efficient action plans based on natural language instructions.However,for embodied tasks,where robots interact with complex environments,textonly LLMs often face challenges due to a lack of compatibility with robotic visual perception.This study provides a comprehensive overview of the emerging integration of LLMs and multimodal LLMs into various robotic tasks.Additionally,we propose a framework that utilizes multimodal GPT-4V to enhance embodied task planning through the combination of natural language instructions and robot visual perceptions.Our results,based on diverse datasets,indicate that GPT-4V effectively enhances robot performance in embodied tasks.This extensive survey and evaluation of LLMs and multimodal LLMs across a variety of robotic tasks enriches the understanding of LLM-centric embodied intelligence and provides forward-looking insights towards bridging the gap in Human-Robot-Environment interaction.
文摘The streamflow over the Yellow River basin is simulated using the PRECIS （Providing REgional Climates for Impacts Studies） regional climate model driven by 15-year （1979-1993） ECMWF reanalysis data as the initial and lateral boundary conditions and an off-line large-scale routing model （LRM）. The LRM uses physical catchment and river channel information and allows streamflow to be predicted for large continental rivers with a 1°×1° spatial resolution. The results show that the PRECIS model can reproduce the general southeast to northwest gradient distribution of the precipitation over the Yellow River basin, The PRECIS- LRM model combination has the capability to simulate the seasonal and annual streamflow over the Yellow River basin. The simulated streamflow is generally coincident with the naturalized streamflow both in timing and in magnitude.
文摘ChatGPT is a powerful artificial intelligence(AI)language model that has demonstrated significant improvements in various natural language processing(NLP) tasks. However, like any technology, it presents potential security risks that need to be carefully evaluated and addressed. In this survey, we provide an overview of the current state of research on security of using ChatGPT, with aspects of bias, disinformation, ethics, misuse,attacks and privacy. We review and discuss the literature on these topics and highlight open research questions and future directions.Through this survey, we aim to contribute to the academic discourse on AI security, enriching the understanding of potential risks and mitigations. We anticipate that this survey will be valuable for various stakeholders involved in AI development and usage, including AI researchers, developers, policy makers, and end-users.
基金supported by the National Key R&D Program of China(No.2021YFB0301200)National Natural Science Foundation of China(No.62025208).
文摘Large-scale Language Models(LLMs)have achieved significant breakthroughs in Natural Language Processing(NLP),driven by the pre-training and fine-tuning paradigm.While this approach allows models to specialize in specific tasks with reduced training costs,the substantial memory requirements during fine-tuning present a barrier to broader deployment.Parameter-Efficient Fine-Tuning(PEFT)techniques,such as Low-Rank Adaptation(LoRA),and parameter quantization methods have emerged as solutions to address these challenges by optimizing memory usage and computational efficiency.Among these,QLoRA,which combines PEFT and quantization,has demonstrated notable success in reducing memory footprints during fine-tuning,prompting the development of various QLoRA variants.Despite these advancements,the quantitative impact of key variables on the fine-tuning performance of quantized LLMs remains underexplored.This study presents a comprehensive analysis of these key variables,focusing on their influence across different layer types and depths within LLM architectures.Our investigation uncovers several critical findings:(1)Larger layers,such as MLP layers,can maintain performance despite reductions in adapter rank,while smaller layers,like self-attention layers,aremore sensitive to such changes;(2)The effectiveness of balancing factors depends more on specific values rather than layer type or depth;(3)In quantization-aware fine-tuning,larger layers can effectively utilize smaller adapters,whereas smaller layers struggle to do so.These insights suggest that layer type is a more significant determinant of fine-tuning success than layer depth when optimizing quantized LLMs.Moreover,for the same discount of trainable parameters,reducing the trainable parameters in a larger layer is more effective in preserving fine-tuning accuracy than in a smaller one.This study provides valuable guidance for more efficient fine-tuning strategies and opens avenues for further research into optimizing LLM fine-tuning in resource-constrained environments.
基金2024 Provincial Teaching Reform Program for Graduate Students in the Second Batch of the 14th Five-Year Plan of Zhejiang Provincial Office of Education:Innovation and Practice of“Six Synergistic”Graduate Teaching Guided by Educator’s Spirit(No.JGCG2024406)Key Project of Zhejiang Provincial Education Science Planning:Research on an interdisciplinary teaching model to promote students’computational thinking from multiple analytical perspectives[No.2025SB103].
文摘Oral expression skills play an essential role in the development of EFL students’language abilities,and how to improve EFL students’oral expression skills is an essential and challenging task.This study adopts a quasi-experimental research method to carry out the research and proposes an AI-based reflective dialogue model.Based on this,an analysis of the impact brought by this model on EFL students’oral expression performance and learning anxiety levels.The results show that students in the experimental group have significantly higher oral expression performance than those in the control group in the three dimensions of grammatical accuracy,expressive fluency,and word accuracy.In addition,the students in the experimental group produced facilitated anxiety after using the AI-based reflective dialogue model for oral expression learning,which prompted the students to learn more diligently.
基金supported by the Fujian Science Foundation for Outstanding Youth(Grant No.2023J06039)the National Natural Science Foundation of China(Grant No.41977259 and No.U2005205)Fujian Province natural resources science and technology innovation project(Grant No.KY-090000-04-2022-019)。
文摘Bedding slope is a typical heterogeneous slope consisting of different soil/rock layers and is likely to slide along the weakest interface.Conventional slope protection methods for bedding slopes,such as retaining walls,stabilizing piles,and anchors,are time-consuming and labor-and energy-intensive.This study proposes an innovative polymer grout method to improve the bearing capacity and reduce the displacement of bedding slopes.A series of large-scale model tests were carried out to verify the effectiveness of polymer grout in protecting bedding slopes.Specifically,load-displacement relationships and failure patterns were analyzed for different testing slopes with various dosages of polymer.Results show the great potential of polymer grout in improving bearing capacity,reducing settlement,and protecting slopes from being crushed under shearing.The polymer-treated slopes remained structurally intact,while the untreated slope exhibited considerable damage when subjected to loads surpassing the bearing capacity.It is also found that polymer-cemented soils concentrate around the injection pipe,forming a fan-shaped sheet-like structure.This study proves the improvement of polymer grouting for bedding slope treatment and will contribute to the development of a fast method to protect bedding slopes from landslides.
基金financially supported by the Open Research Fund of Hunan Provincial Key Laboratory of Key Technology on Hydropower Development (Grant No.PKLHD202003)the National Natural Science Foundation of China (Grant Nos.52071058 and 51939002)+1 种基金the National Natural Science Foundation of Liaoning Province (Grant No.2022-KF-18-01)Fundamental Research Funds for the Central University (Grant No.DUT20ZD219)。
文摘Considering the large diameter effect of piles,the influence of different pile-soil analysis methods on the design of monopile foundations for offshore wind turbines has become an urgent problem to be solved.Three different pile-soil models were used to study a large 10 MW monopile wind turbine.By modeling the three models in the SACS software,this paper analyzed the motion response of the overall structure under the conditions of wind and waves.According to the given working conditions,this paper concludes that under the condition of independent wind,the average value of the tower top x-displacement of the rigid connection method is the smalle st,and the standard deviation is the smallest under the condition of independent wave.The results obtained by the p-y curve method are the most conservative.
基金supported by the National Key R&D Program of China with Grant number 2019YFB1803400the National Natural Science Foundation of China under Grant number 62071114the Fundamental Research Funds for the Central Universities of China under grant numbers 3204002004A2 and 2242022k30005。
文摘This paper investigates the wireless communication with a novel architecture of antenna arrays,termed modular extremely large-scale array(XLarray),where array elements of an extremely large number/size are regularly mounted on a shared platform with both horizontally and vertically interlaced modules.Each module consists of a moderate/flexible number of array elements with the inter-element distance typically in the order of the signal wavelength,while different modules are separated by the relatively large inter-module distance for convenience of practical deployment.By accurately modelling the signal amplitudes and phases,as well as projected apertures across all modular elements,we analyse the near-field signal-to-noise ratio(SNR)performance for modular XL-array communications.Based on the non-uniform spherical wave(NUSW)modelling,the closed-form SNR expression is derived in terms of key system parameters,such as the overall modular array size,distances of adjacent modules along all dimensions,and the user's three-dimensional(3D)location.In addition,with the number of modules in different dimensions increasing infinitely,the asymptotic SNR scaling laws are revealed.Furthermore,we show that our proposed near-field modelling and performance analysis include the results for existing array architectures/modelling as special cases,e.g.,the collocated XL-array architecture,the uniform plane wave(UPW)based far-field modelling,and the modular extremely large-scale uniform linear array(XL-ULA)of onedimension.Extensive simulation results are presented to validate our findings.
