Comparative analysis of empirical and deep learning models for ionospheric sporadic E layer prediction

Sporadic E(Es)layers in the ionosphere are characterized by intense plasma irregularities in the E region at altitudes of 90-130 km.Because they can significantly influence radio communications and navigation systems,accurate forecasting of Es layers is crucial for ensuring the precision and dependability of navigation satellite systems.In this study,we present Es predictions made by an empirical model and by a deep learning model,and analyze their differences comprehensively by comparing the model predictions to satellite RO measurements and ground-based ionosonde observations.The deep learning model exhibited significantly better performance,as indicated by its high coefficient of correlation(r=0.87)with RO observations and predictions,than did the empirical model(r=0.53).This study highlights the importance of integrating artificial intelligence technology into ionosphere modelling generally,and into predicting Es layer occurrences and characteristics,in particular.

Dynamics of a Stochastic Epidemic Model with Age-group

A stochastic epidemic model with two age groups is established in this study,in which the susceptible(S),the exposed(E),the infected(I),the hospitalized(H)and the recovered(R)are involved within the total population,the aging rates between two age groups are set to be constant.The existence-and-uniqueness of global positive solution is firstly showed.Then,by constructing several appropriate Lyapunov functions and using the high-dimensional Itô’s formula,the sufficient conditions for the stochastic extinction and stochastic persistence of the exposed individuals and the infected individuals are obtained.The stochastic extinction indicator and the stochastic persistence indicator are less-valued expressions compared with the basic reproduction number.Meanwhile,the main results of this study are modified into multi-age groups.Furthermore,by using the surveillance data for Fujian Provincial Center for Disease Control and Prevention,Fuzhou COVID-19 epidemic is chosen to carry out the numerical simulations,which show that the age group of the population plays the vital role when studying infectious diseases.

Aquaporin-4-IgG-seropositive neuromyelitis optica spectrum disorders:progress of experimental models based on disease pathogenesis

Neuromyelitis optica spectrum disorders are neuroinflammatory demyelinating disorders that lead to permanent visual loss and motor dysfunction.To date,no effective treatment exists as the exact causative mechanism remains unknown.Therefore,experimental models of neuromyelitis optica spectrum disorders are essential for exploring its pathogenesis and in screening for therapeutic targets.Since most patients with neuromyelitis optica spectrum disorders are seropositive for IgG autoantibodies against aquaporin-4,which is highly expressed on the membrane of astrocyte endfeet,most current experimental models are based on aquaporin-4-IgG that initially targets astrocytes.These experimental models have successfully simulated many pathological features of neuromyelitis optica spectrum disorders,such as aquaporin-4 loss,astrocytopathy,granulocyte and macrophage infiltration,complement activation,demyelination,and neuronal loss;however,they do not fully capture the pathological process of human neuromyelitis optica spectrum disorders.In this review,we summarize the currently known pathogenic mechanisms and the development of associated experimental models in vitro,ex vivo,and in vivo for neuromyelitis optica spectrum disorders,suggest potential pathogenic mechanisms for further investigation,and provide guidance on experimental model choices.In addition,this review summarizes the latest information on pathologies and therapies for neuromyelitis optica spectrum disorders based on experimental models of aquaporin-4-IgG-seropositive neuromyelitis optica spectrum disorders,offering further therapeutic targets and a theoretical basis for clinical trials.

Solar flare forecasting based on a Fusion Model

Solar flare prediction is an important subject in the field of space weather.Deep learning technology has greatly promoted the development of this subject.In this study,we propose a novel solar flare forecasting model integrating Deep Residual Network(ResNet)and Support Vector Machine(SVM)for both≥C-class(C,M,and X classes)and≥M-class(M and X classes)flares.We collected samples of magnetograms from May 1,2010 to September 13,2018 from Space-weather Helioseismic and Magnetic Imager(HMI)Active Region Patches and then used a cross-validation method to obtain seven independent data sets.We then utilized five metrics to evaluate our fusion model,based on intermediate-output extracted by ResNet and SVM using the Gaussian kernel function.Our results show that the primary metric true skill statistics(TSS)achieves a value of 0.708±0.027 for≥C-class prediction,and of 0.758±0.042 for≥M-class prediction;these values indicate that our approach performs significantly better than those of previous studies.The metrics of our fusion model’s performance on the seven datasets indicate that the model is quite stable and robust,suggesting that fusion models that integrate an excellent baseline network with SVM can achieve improved performance in solar flare prediction.Besides,we also discuss the performance impact of architectural innovation in our fusion model.

Engine Misfire Fault Detection Based on the Channel Attention Convolutional Model

To accurately diagnosemisfire faults in automotive engines,we propose a Channel Attention Convolutional Model,specifically the Squeeze-and-Excitation Networks(SENET),for classifying engine vibration signals and precisely pinpointing misfire faults.In the experiment,we established a total of 11 distinct states,encompassing the engine’s normal state,single-cylinder misfire faults,and dual-cylinder misfire faults for different cylinders.Data collection was facilitated by a highly sensitive acceleration signal collector with a high sampling rate of 20,840Hz.The collected data were methodically divided into training and testing sets based on different experimental groups to ensure generalization and prevent overlap between the two sets.The results revealed that,with a vibration acceleration sequence of 1000 time steps(approximately 50 ms)as input,the SENET model achieved a misfire fault detection accuracy of 99.8%.For comparison,we also trained and tested several commonly used models,including Long Short-Term Memory(LSTM),Transformer,and Multi-Scale Residual Networks(MSRESNET),yielding accuracy rates of 84%,79%,and 95%,respectively.This underscores the superior accuracy of the SENET model in detecting engine misfire faults compared to other models.Furthermore,the F1 scores for each type of recognition in the SENET model surpassed 0.98,outperforming the baseline models.Our analysis indicated that the misclassified samples in the LSTM and Transformer models’predictions were primarily due to intra-class misidentifications between single-cylinder and dual-cylinder misfire scenarios.To delve deeper,we conducted a visual analysis of the features extracted by the LSTM and SENET models using T-distributed Stochastic Neighbor Embedding(T-SNE)technology.The findings revealed that,in the LSTMmodel,data points of the same type tended to cluster together with significant overlap.Conversely,in the SENET model,data points of various types were more widely and evenly dispersed,demonstrating its effectiveness in distinguishing between different fault types.

Optimizing Fine-Tuning in Quantized Language Models:An In-Depth Analysis of Key Variables

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.

Exploiting fly models to investigate rare human neurological disorders

Rare neurological diseases,while individually are rare,collectively impact millions globally,leading to diverse and often severe neurological symptoms.Often attributed to genetic mutations that disrupt protein function or structure,understanding their genetic basis is crucial for accurate diagnosis and targeted therapies.To investigate the underlying pathogenesis of these conditions,researchers often use non-mammalian model organisms,such as Drosophila(fruit flies),which is valued for their genetic manipulability,cost-efficiency,and preservation of genes and biological functions across evolutionary time.Genetic tools available in Drosophila,including CRISPR-Cas9,offer a means to manipulate gene expression,allowing for a deep exploration of the genetic underpinnings of rare neurological diseases.Drosophila boasts a versatile genetic toolkit,rapid generation turnover,and ease of large-scale experimentation,making it an invaluable resource for identifying potential drug candidates.Researchers can expose flies carrying disease-associated mutations to various compounds,rapidly pinpointing promising therapeutic agents for further investigation in mammalian models and,ultimately,clinical trials.In this comprehensive review,we explore rare neurological diseases where fly research has significantly contributed to our understanding of their genetic basis,pathophysiology,and potential therapeutic implications.We discuss rare diseases associated with both neuron-expressed and glial-expressed genes.Specific cases include mutations in CDK19 resulting in epilepsy and developmental delay,mutations in TIAM1 leading to a neurodevelopmental disorder with seizures and language delay,and mutations in IRF2BPL causing seizures,a neurodevelopmental disorder with regression,loss of speech,and abnormal movements.And we explore mutations in EMC1 related to cerebellar atrophy,visual impairment,psychomotor retardation,and gain-of-function mutations in ACOX1 causing Mitchell syndrome.Loss-of-function mutations in ACOX1 result in ACOX1 deficiency,characterized by very-long-chain fatty acid accumulation and glial degeneration.Notably,this review highlights how modeling these diseases in Drosophila has provided valuable insights into their pathophysiology,offering a platform for the rapid identification of potential therapeutic interventions.Rare neurological diseases involve a wide range of expression systems,and sometimes common phenotypes can be found among different genes that cause abnormalities in neurons or glia.Furthermore,mutations within the same gene may result in varying functional outcomes,such as complete loss of function,partial loss of function,or gain-of-function mutations.The phenotypes observed in patients can differ significantly,underscoring the complexity of these conditions.In conclusion,Drosophila represents an indispensable and cost-effective tool for investigating rare neurological diseases.By facilitating the modeling of these conditions,Drosophila contributes to a deeper understanding of their genetic basis,pathophysiology,and potential therapies.This approach accelerates the discovery of promising drug candidates,ultimately benefiting patients affected by these complex and understudied diseases.

A promising approach for quantifying focal stroke modeling and assessing stroke progression:optical resolution photoacoustic microscopy photothrombosis

To investigate the mechanisms underlying the onset and progression of ischemic stroke,some methods have been proposed that can simultaneously monitor and create embolisms in the animal cerebral cortex.However,these methods often require complex systems and the effect of age on cerebral embolism has not been adequately studied,although ischemic stroke is strongly age-related.In this study,we propose an optical-resolution photoacoustic microscopy-based visualized photothrombosis methodology to create and monitor ischemic stroke in mice simultaneously using a 532 nm pulsed laser.We observed the molding process in mice of different ages and presented age-dependent vascular embolism differentiation.Moreover,we integrated optical coherence tomography angiography to investigate age-associated trends in cerebrovascular variability following a stroke.Our imaging data and quantitative analyses underscore the differential cerebrovascular responses to stroke in mice of different ages,thereby highlighting the technique's potential for evaluating cerebrovascular health and unraveling age-related mechanisms involved in ischemic strokes.

Vertical gradients of neutral winds observed by ICON and estimated by the Horizontal Wind Model during the geomagnetic storm on August 26−28,2021

The Michelson Interferometer for Global High-resolution Thermospheric Imaging(MIGHTI)onboard the Ionospheric Connection Explorer(ICON)satellite offers the opportunity to investigate the altitude profile of thermospheric winds.In this study,we used the red-line measurements of MIGHTI to compare with the results estimated by Horizontal Wind Model 14(HWM14).The data selected included both the geomagnetic quiet period(December 2019 to August 2022)and the geomagnetic storm on August 26-28,2021.During the geomagnetic quiet period,the estimations of neutral winds from HWM14 showed relatively good agreement with the observations from ICON.According to the ICON observations,near the equator,zonal winds reverse from westward to eastward at around 06:00 local time(LT)at higher altitudes,and the stronger westward winds appear at later LTs at lower altitudes.At around 16:00 LT,eastward winds at 300 km reverse to westward,and vertical gradients of zonal winds similar to those at sunrise hours can be observed.In the middle latitudes,zonal winds reverse about 2-4 h earlier.Meridional winds vary more significantly than zonal winds with seasonal and latitudinal variations.According to the ICON observations,in the northern low latitudes,vertical reversals of meridional winds are found at 08:00-13:00 LT from 300 to 160 km and at around 18:00 LT from 300 to 200 km during the June solstice.Similar reversals of meridional winds are found at 04:00-07:00 LT from 300 to 160 km and at 22:00-02:00 LT from 270 to 200 km during the December solstice.In the southern low latitudes,meridional wind reversals occur at 08:00-11:00 LT from 200 to 160 km and at 21:00-02:00 LT from 300 to 200 km during the June solstice.During the December solstice,reversals of the meridional wind appear at 20:00-01:00 LT below 200 km and at 06:00-11:00 LT from 300 to 160 km.In the northern middle latitudes,the northward winds are dominant at 08:00-14:00 LT at 230 km during the June solstice.Northward winds persist until 16:00 LT at 160 and 300 km.During the December solstice,the northward winds are dominant from 06:00 to 21:00 LT.The vertical variations in neutral winds during the geomagnetic storm on August 26-28 were analyzed in detail.Both meridional and zonal winds during the active geomagnetic period observed by ICON show distinguishable vertical shear structures at different stages of the storm.On the dayside,during the main phase,the peak velocities of westward winds extend from a higher altitude to a lower altitude,whereas during the recovery phase,the peak velocities of the westward winds extend from lower altitudes to higher altitudes.The velocities of the southward winds are stronger at lower altitudes during the storm.These vertical structures of horizontal winds during the storm could not be reproduced by the HWM14 wind estimations,and the overall response to the storm of the horizontal winds in the low and middle latitudes is underestimated by HWM14.The ICON observations provide a good dataset for improving the HWM wind estimations in the middle and upper atmosphere,especially the vertical variations.

Reduced mesencephalic astrocyte-derived neurotrophic factor expression by mutant androgen receptor contributes to neurodegeneration in a model of spinal and bulbar muscular atrophy pathology

Spinal and bulbar muscular atrophy is a neurodegenerative disease caused by extended CAG trinucleotide repeats in the androgen receptor gene,which encodes a ligand-dependent transcription facto r.The mutant androgen receptor protein,characterized by polyglutamine expansion,is prone to misfolding and forms aggregates in both the nucleus and cytoplasm in the brain in spinal and bulbar muscular atrophy patients.These aggregates alter protein-protein interactions and compromise transcriptional activity.In this study,we reported that in both cultured N2a cells and mouse brain,mutant androgen receptor with polyglutamine expansion causes reduced expression of mesencephalic astrocyte-de rived neurotrophic factor.Overexpressio n of mesencephalic astrocyte-derived neurotrophic factor amelio rated the neurotoxicity of mutant androgen receptor through the inhibition of mutant androgen receptor aggregation.Conversely.knocking down endogenous mesencephalic astrocyte-derived neurotrophic factor in the mouse brain exacerbated neuronal damage and mutant androgen receptor aggregation.Our findings suggest that inhibition of mesencephalic astrocyte-derived neurotrophic factor expression by mutant androgen receptor is a potential mechanism underlying neurodegeneration in spinal and bulbar muscular atrophy.

Assessing the possibility of using large language models in ocular surface diseases

AIM:To assess the possibility of using different large language models(LLMs)in ocular surface diseases by selecting five different LLMS to test their accuracy in answering specialized questions related to ocular surface diseases:ChatGPT-4,ChatGPT-3.5,Claude 2,PaLM2,and SenseNova.METHODS:A group of experienced ophthalmology professors were asked to develop a 100-question singlechoice question on ocular surface diseases designed to assess the performance of LLMs and human participants in answering ophthalmology specialty exam questions.The exam includes questions on the following topics:keratitis disease(20 questions),keratoconus,keratomalaciac,corneal dystrophy,corneal degeneration,erosive corneal ulcers,and corneal lesions associated with systemic diseases(20 questions),conjunctivitis disease(20 questions),trachoma,pterygoid and conjunctival tumor diseases(20 questions),and dry eye disease(20 questions).Then the total score of each LLMs and compared their mean score,mean correlation,variance,and confidence were calculated.RESULTS:GPT-4 exhibited the highest performance in terms of LLMs.Comparing the average scores of the LLMs group with the four human groups,chief physician,attending physician,regular trainee,and graduate student,it was found that except for ChatGPT-4,the total score of the rest of the LLMs is lower than that of the graduate student group,which had the lowest score in the human group.Both ChatGPT-4 and PaLM2 were more likely to give exact and correct answers,giving very little chance of an incorrect answer.ChatGPT-4 showed higher credibility when answering questions,with a success rate of 59%,but gave the wrong answer to the question 28% of the time.CONCLUSION:GPT-4 model exhibits excellent performance in both answer relevance and confidence.PaLM2 shows a positive correlation(up to 0.8)in terms of answer accuracy during the exam.In terms of answer confidence,PaLM2 is second only to GPT4 and surpasses Claude 2,SenseNova,and GPT-3.5.Despite the fact that ocular surface disease is a highly specialized discipline,GPT-4 still exhibits superior performance,suggesting that its potential and ability to be applied in this field is enormous,perhaps with the potential to be a valuable resource for medical students and clinicians in the future.

采用STAMP-24Model的多组织事故分析

安全生产事故往往由多组织交互、多因素耦合造成,事故原因涉及多个组织。为预防和遏制多组织生产安全事故的发生,基于系统理论事故建模与过程模型(Systems-Theory Accident Modeling and Process,STAMP)、24Model,构建一种用于多组织事故分析的方法,并以青岛石油爆炸事故为例进行事故原因分析。结果显示:STAMP-24Model可以分组织,分层次且有效、全面、详细地分析涉及多个组织的事故原因,探究多组织之间的交互关系;对事故进行动态演化分析,可得到各组织不安全动作耦合关系与形成的事故失效链及管控失效路径,进而为预防多组织事故提供思路和参考。

基于改进24Model-ISM-SNA建筑工人不安全行为关联路径研究

建筑施工现场环境复杂,为有效控制不安全行为发生,基于行为安全“2-4”模型对360份具有代表性的建筑安全事故调查报告进行分析,提取出22个不安全行为的主要影响因素。利用灰色关联分析方法(GRA)改进的集成ISM-SNA模型,将不安全行为风险因素划分为表层、过渡层与深层,然后对风险因素进行可视化分析、中心度分析及凝聚子群分析,揭示了各致因因素间的关联关系和传导路径。结果表明,建筑工人不安全行为影响因素可划分成7级3阶的多级递阶结构,安全意识、现场监管、外部环境是建筑工人不安全行为的关键影响因素,同时现场监管和隐患排查到位能有效降低不安全行为的发生。

基于24Model的地铁内涝事故原因分析与评估

为降低地铁内涝事故灾害风险,基于事故致因“2-4”模型,分析了地铁内涝事故致灾因子,采用层次分析法构建了地铁内涝事故原因分析指标体系,确定了各风险因子的权重,并利用模糊综合评价法对地铁内涝事故进行了定量评估,识别出关键的影响因素。结果表明:地铁内涝事故一级指标中不安全动作与物态因素最重要,其中影响最大的指标包括擅自更改建筑设计、未按照要求检查水位情况、未及时排水、出入口不符合防汛标准等因素;习惯性不安全行为的权重位居第二,表明该指标因素较为重要,同时安全管理体系得分位居第二,表明该指标因素较易发生。对关键指标采取防范措施,可有效降低风险,从而减少地铁内涝事故的发生。

基于24Model-D-ISM的地铁站火灾疏散影响因素研究

为预防地铁站火灾事故,深入了解地铁站火灾人员疏散影响因素间的内在联系与层次结构,基于第6版“2-4”模型(24Model)分析63起地铁站火灾疏散事故,充分考虑各个因素之间的交互作用,提取19个影响地铁站人员疏散的关键因素,建立地铁站火灾人员疏散影响因素指标体系;采用算子客观赋权法(C-OWA)改进决策试验与评价实验法(DEMATEL),确定地铁站火灾人员疏散的重要影响因素;在此基础上,采用解释结构模型(ISM)分析各个因素间的层次结构及相互作用路径,构建地铁站火灾人员疏散影响因素的多级递阶结构模型。研究结果表明:疏散引导、恐慌从众行为、人员拥挤为地铁站火灾人员疏散的关键影响因素;地铁站火灾人员疏散受表层因素、中间层因素、深层因素共同作用的影响,其中,疏散教育与培训、设施维护与检查、疏散预案等因素是根源影响因素,重视根源影响因素的改善有利于从本质上预防和控制事故的发生。

24Model与LCM原因因素定义对比研究

为探究损失致因模型(LCM)原因因素定义与事故致因“2-4”模型(24Model)存在的异同和优缺点,梳理2个模型各层面原因和结果的定义,对比定义内容及其对事故原因分析等安全实务的指导作用,并以一起瓦斯爆炸事故为例加以实证分析,获得二者分析结果之间的差异。研究结果表明:LCM是首个将管理因素纳入事故致因分析的一维事件序列模型,可明确各层面原因因素的定义和因素间的逻辑关系,但部分定义存在交叉重复的问题,并没有揭示安全工作指导思想等深层次事故致因因素;24Model作为系统性事故致因模型,对各类因素的定义均以组织为主体,描述事件、事故、安全的概念内涵,划分个体安全动作、安全能力和组织安全管理体系的类别并给出含义解析,探究组织安全文化层面的问题并以32个元素体现;2个模型的事故原因分析方法均建立在对各层级原因因素定义的基础上,并适用于模型理论体系本身。

EXPERIMENTS OF A REDUCED GRID IN LASG/IAP WORLD OCEAN GENERAL CIRCULATION MODELS (OGCMs)

Due to the decrease in grid size associated with the convergence of meridians toward the poles inspherical coordinates, the time steps in many global climate models with finite-difference method are restrictedto be unpleasantly small. To overcome the problem, a reduced grid is introduced to LASG/IAP world oceangeneral circulation models. The reduced grid is implemented successfully in the coarser resolutions versionmodel L30T63 at first. Then, it is carried out in the improved version model LICOM with finer resolutions. Inthe experiment with model L30T63, under time step unchanged though, execution time per single model run isshortened significantly owing to the decrease of grid number and filtering execution in high latitudes. Resultsfrom additional experiments with L30T63 show that the time step of integration can be quadrupled at most inreduced grid with refinement ratio 3. In the experiment with model LICOM and with the model’s original timestep unchanged, the model covered area is extended to the whole globe from its original case with the grid pointof North Pole considered as an isolated island and the results of experiment are shown to be acceptable.

Anisotropic time-dependent behaviors of shale under direct shearing and associated empirical creep models

Understanding the anisotropic creep behaviors of shale under direct shearing is a challenging issue.In this context,we conducted shear-creep and steady-creep tests on shale with five bedding orientations (i.e.0°,30°,45°,60°,and 90°),under multiple levels of direct shearing for the first time.The results show that the anisotropic creep of shale exhibits a significant stress-dependent behavior.Under a low shear stress,the creep compliance of shale increases linearly with the logarithm of time at all bedding orientations,and the increase depends on the bedding orientation and creep time.Under high shear stress conditions,the creep compliance of shale is minimal when the bedding orientation is 0°,and the steady-creep rate of shale increases significantly with increasing bedding orientations of 30°,45°,60°,and 90°.The stress-strain values corresponding to the inception of the accelerated creep stage show an increasing and then decreasing trend with the bedding orientation.A semilogarithmic model that could reflect the stress dependence of the steady-creep rate while considering the hardening and damage process is proposed.The model minimizes the deviation of the calculated steady-state creep rate from the observed value and reveals the behavior of the bedding orientation's influence on the steady-creep rate.The applicability of the five classical empirical creep models is quantitatively evaluated.It shows that the logarithmic model can well explain the experimental creep strain and creep rate,and it can accurately predict long-term shear creep deformation.Based on an improved logarithmic model,the variations in creep parameters with shear stress and bedding orientations are discussed.With abovementioned findings,a mathematical method for constructing an anisotropic shear creep model of shale is proposed,which can characterize the nonlinear dependence of the anisotropic shear creep behavior of shale on the bedding orientation.

Projecting Wintertime Newly Formed Arctic Sea Ice through Weighting CMIP6 Model Performance and Independence

Precipitous Arctic sea-ice decline and the corresponding increase in Arctic open-water areas in summer months give more space for sea-ice growth in the subsequent cold seasons. Compared to the decline of the entire Arctic multiyear sea ice,changes in newly formed sea ice indicate more thermodynamic and dynamic information on Arctic atmosphere–ocean–ice interaction and northern mid–high latitude atmospheric teleconnections. Here, we use a large multimodel ensemble from phase 6 of the Coupled Model Intercomparison Project(CMIP6) to investigate future changes in wintertime newly formed Arctic sea ice. The commonly used model-democracy approach that gives equal weight to each model essentially assumes that all models are independent and equally plausible, which contradicts with the fact that there are large interdependencies in the ensemble and discrepancies in models' performances in reproducing observations. Therefore, instead of using the arithmetic mean of well-performing models or all available models for projections like in previous studies, we employ a newly developed model weighting scheme that weights all models in the ensemble with consideration of their performance and independence to provide more reliable projections. Model democracy leads to evident bias and large intermodel spread in CMIP6 projections of newly formed Arctic sea ice. However, we show that both the bias and the intermodel spread can be effectively reduced by the weighting scheme. Projections from the weighted models indicate that wintertime newly formed Arctic sea ice is likely to increase dramatically until the middle of this century regardless of the emissions scenario.Thereafter, it may decrease(or remain stable) if the Arctic warming crosses a threshold(or is extensively constrained).

Evolution and Prospects of Foundation Models: From Large Language Models to Large Multimodal Models

Since the 1950s,when the Turing Test was introduced,there has been notable progress in machine language intelligence.Language modeling,crucial for AI development,has evolved from statistical to neural models over the last two decades.Recently,transformer-based Pre-trained Language Models(PLM)have excelled in Natural Language Processing(NLP)tasks by leveraging large-scale training corpora.Increasing the scale of these models enhances performance significantly,introducing abilities like context learning that smaller models lack.The advancement in Large Language Models,exemplified by the development of ChatGPT,has made significant impacts both academically and industrially,capturing widespread societal interest.This survey provides an overview of the development and prospects from Large Language Models(LLM)to Large Multimodal Models(LMM).It first discusses the contributions and technological advancements of LLMs in the field of natural language processing,especially in text generation and language understanding.Then,it turns to the discussion of LMMs,which integrates various data modalities such as text,images,and sound,demonstrating advanced capabilities in understanding and generating cross-modal content,paving new pathways for the adaptability and flexibility of AI systems.Finally,the survey highlights the prospects of LMMs in terms of technological development and application potential,while also pointing out challenges in data integration,cross-modal understanding accuracy,providing a comprehensive perspective on the latest developments in this field.