With the continuous evolution and expanding applications of Large Language Models (LLMs), there has been a noticeable surge in the size of the emerging models. It is not solely the growth in model size, primarily meas...With the continuous evolution and expanding applications of Large Language Models (LLMs), there has been a noticeable surge in the size of the emerging models. It is not solely the growth in model size, primarily measured by the number of parameters, but also the subsequent escalation in computational demands, hardware and software prerequisites for training, all culminating in a substantial financial investment as well. In this paper, we present novel techniques like supervision, parallelization, and scoring functions to get better results out of chains of smaller language models, rather than relying solely on scaling up model size. Firstly, we propose an approach to quantify the performance of a Smaller Language Models (SLM) by introducing a corresponding supervisor model that incrementally corrects the encountered errors. Secondly, we propose an approach to utilize two smaller language models (in a network) performing the same task and retrieving the best relevant output from the two, ensuring peak performance for a specific task. Experimental evaluations establish the quantitative accuracy improvements on financial reasoning and arithmetic calculation tasks from utilizing techniques like supervisor models (in a network of model scenario), threshold scoring and parallel processing over a baseline study.展开更多
AIM:To compare the imaging results with histology and to evaluate the diagnostic sensitivity of imaging modalities for hepatocellular carcinoma(HCC)smaller than 2 cm.METHODS:Nodules smaller than 2 cm(n=34)revealed by ...AIM:To compare the imaging results with histology and to evaluate the diagnostic sensitivity of imaging modalities for hepatocellular carcinoma(HCC)smaller than 2 cm.METHODS:Nodules smaller than 2 cm(n=34)revealed by ultrasonography(US)in 29 patients with liver cirrhosis were analyzed.Histological diagnosis of HCC was performed by ultrasonographic guidance:moderately-differentiated HCC(n=24);well-differentiated HCC(n=10).The patterns disclosed by the four imaging modalities defined the conclusive diagnosis of HCC:(1)contrast-enhanced computed tomography(CECT),hypervascularity in the arterial phase and washout in the equilibrium phase;(2)Sonazoid contrast-enhanced US(CEUS),hypervascularity in the early vascular phase and defect in the Kupffer phase;(3)gadolinium-ethoxybenzyl-diethylenetriamine pentaacetic acid(Gd-EOBDTPA)-enhanced magnetic resonance imaging(MRI),hypervascularity in the arterial phase and/or defect in the hepatobiliary phase;and(4)CT arterioportal angiography:hypervascularity by CT during arteriography and/ or perfusion defect by CT during arterial portography.RESULTS:Overall,the sensitivity of diagnosing HCC smaller than 2 cm was 52.9%(18/34)(95%CI:35.170.2)by CECT;67.6%(23/34)(95%CI:49.5-82.6)by Sonazoid CEUS;76.5%(26/34)(95%CI:58.8-89.3) by Gd-EOB-DTPA MRI;and 88.2%(30/34)(95%CI: 72.5-96.7)by CT arterioportal angiography.The diagnostic sensitivity of detecting moderately-differentiated HCC by CECT,Sonazoid CEUS,Gd-EOB-DTPA MRI and CT arterioportal angiography was 62.5%(15/24)(95%CI: 40.6-81.2),79.2%(19/24)(95%CI:57.8-92.9),75.0% (18/24)(95%CI:53.3-90.2)and 95.8%(23/24)(95% CI:78.9-99.9),respectively.A significant difference(P< 0.05)was observed between CECT and CT arterioportal angiography in all nodules.There was no difference between Sonazoid CEUS,Gd-EOB-DTPA MRI,and CT arterioportal angiography.The combined sensitivity of Sonazoid CEUS and Gd-EOB-DTPA MRI was 94.1%(32/34).CONCLUSION:Changing the main diagnostic modality for HCC smaller than 2 cm from CT arterioportal angiography to Sonazoid CEUS and Gd-EOB-DTPA MRI is recommended.展开更多
Smaller mesoscale eddies(SMEs)have an important effect on the transmission of ocean temperatures,salinity,energy,and marine biochemical processes.However,traditional altimeters,the dominant sensors used to identify an...Smaller mesoscale eddies(SMEs)have an important effect on the transmission of ocean temperatures,salinity,energy,and marine biochemical processes.However,traditional altimeters,the dominant sensors used to identify and track eddies,have made it challenging to observe SMEs accurately due to resolution limitations.Eddies drive local upwelling or downwelling,leaving signatures on sea surface temperatures(SSTs)and chlorophyll concentrations(Chls).SST can be observed by spaceborne infrared sensors,and Chl can be measured by ocean color remote sensing.Therefore,multisatellite observations provide an opportunity to obtain information to characterize SMEs.In this paper,an eddy detection algorithm based on SST and Chl images is proposed,which identifies eddies by characterizing the spatial and temporal distribution of SST and Chl data.The algorithm is applied to characterize and analyze SMEs in the Kuroshio Extension.Statistical results on their distribution and seasonal variability are shown,and the formation processes are preliminarily discussed.SMEs generation may be contributed by horizontal strain instability,the interaction of topographic obstacles and currents,and wind stress curl.展开更多
China’s big spenders now far-flung by Yu Nan WIDELY recognized as one of the most beautiful cities in China,the coastal port of Dalian is a perfect example of a lowertier city becoming a growth hub for multinationals...China’s big spenders now far-flung by Yu Nan WIDELY recognized as one of the most beautiful cities in China,the coastal port of Dalian is a perfect example of a lowertier city becoming a growth hub for multinationals, owing to stronger purchasing power. That trend inspired Boston Consulting Group(BCG) to release a report,advising businessmen to look beyond the familiar mega-cities to the fast-growing smaller展开更多
“Smaller is softer”is a reverse size dependence of strength,defying the“smaller is stronger”tenet.It usually results from surface-mediated displacive or diffusive deformation and is mainly found in some ultra-smal...“Smaller is softer”is a reverse size dependence of strength,defying the“smaller is stronger”tenet.It usually results from surface-mediated displacive or diffusive deformation and is mainly found in some ultra-small-scale(below tens of nanometers)metallic materials.Here,making use of the surface modifi-cation via ion beam irradiation,we bring the“smaller is softer”into being in a covalently-bonded,hard,and brittle material-amorphous Si(a-Si)at a much larger size regime(<∼500 nm).It is manifested as the transition from the quasi-brittle failure to the homogeneous plastic deformation as well as the de-creasing yield stress with sample volume reduction at the submicron-scale regime.An analytical model of hard core/superplastic shell has been proposed to explain the artificially-controllable size-dependent softening.This surface engineering pathway via ion irradiation is not only of particular interest to tai-lor the strength and deformation behaviors in small-sized a-Si or other covalently-bonded amorphous solids but also of practical relevance to the utility of a-Si in microelectronics and microelectromechanical systems.展开更多
Brain tumors are potentially fatal presence of cancer cells over a human brain,and they need to be segmented for accurate and reliable planning of diag-nosis.Segmentation process must be carried out in different regio...Brain tumors are potentially fatal presence of cancer cells over a human brain,and they need to be segmented for accurate and reliable planning of diag-nosis.Segmentation process must be carried out in different regions based on which the stages of cancer can be accurately derived.Glioma patients exhibit a different level of challenge in terms of cancer or tumors detection as the Magnetic Resonance Imaging(MRI)images possess varying sizes,shapes,positions,and modalities.The scanner used for sensing the location of tumors cells will be sub-jected to additional protocols and measures for accuracy,in turn,increasing the time and affecting the performance of the entire model.In this view,Convolutional Neural Networks deliver suitable models for efficient segmentation and thus delivered promising results.The previous strategies and models failed to adhere to diversity of sizes and shapes,proving to be a well-established solution for detecting tumors of bigger size.Tumors tend to be smaller in size and shape during their premature stages and they can easily evade the algorithms of Convolutional Neural Network(CNN).This proposal intends to furnish a detailed model for sensing early stages of cancer and hence perform segmentation irrespective of the current size and shape of tumors.The size of networks and layers will lead to a significant weightage when multiple kernel sizes are involved,especially in multi-resolution environments.On the other hand,the proposed model is designed with a novel approach including a dilated convolution and level-based learning strat-egy.When the convolution process is dilated,the process of feature extraction deals with multiscale objective and level-based learning eliminates the shortcoming of previous models,thereby enhancing the quality of smaller tumors cells and shapes.The level-based learning approach also encapsulates the feature recon-struction processes which highlights the sensing of small-scale tumors growth.Inclusively,segmenting the images is performed with better accuracy and hence detection becomes better when compared to that of hierarchical approaches.展开更多
文摘With the continuous evolution and expanding applications of Large Language Models (LLMs), there has been a noticeable surge in the size of the emerging models. It is not solely the growth in model size, primarily measured by the number of parameters, but also the subsequent escalation in computational demands, hardware and software prerequisites for training, all culminating in a substantial financial investment as well. In this paper, we present novel techniques like supervision, parallelization, and scoring functions to get better results out of chains of smaller language models, rather than relying solely on scaling up model size. Firstly, we propose an approach to quantify the performance of a Smaller Language Models (SLM) by introducing a corresponding supervisor model that incrementally corrects the encountered errors. Secondly, we propose an approach to utilize two smaller language models (in a network) performing the same task and retrieving the best relevant output from the two, ensuring peak performance for a specific task. Experimental evaluations establish the quantitative accuracy improvements on financial reasoning and arithmetic calculation tasks from utilizing techniques like supervisor models (in a network of model scenario), threshold scoring and parallel processing over a baseline study.
文摘AIM:To compare the imaging results with histology and to evaluate the diagnostic sensitivity of imaging modalities for hepatocellular carcinoma(HCC)smaller than 2 cm.METHODS:Nodules smaller than 2 cm(n=34)revealed by ultrasonography(US)in 29 patients with liver cirrhosis were analyzed.Histological diagnosis of HCC was performed by ultrasonographic guidance:moderately-differentiated HCC(n=24);well-differentiated HCC(n=10).The patterns disclosed by the four imaging modalities defined the conclusive diagnosis of HCC:(1)contrast-enhanced computed tomography(CECT),hypervascularity in the arterial phase and washout in the equilibrium phase;(2)Sonazoid contrast-enhanced US(CEUS),hypervascularity in the early vascular phase and defect in the Kupffer phase;(3)gadolinium-ethoxybenzyl-diethylenetriamine pentaacetic acid(Gd-EOBDTPA)-enhanced magnetic resonance imaging(MRI),hypervascularity in the arterial phase and/or defect in the hepatobiliary phase;and(4)CT arterioportal angiography:hypervascularity by CT during arteriography and/ or perfusion defect by CT during arterial portography.RESULTS:Overall,the sensitivity of diagnosing HCC smaller than 2 cm was 52.9%(18/34)(95%CI:35.170.2)by CECT;67.6%(23/34)(95%CI:49.5-82.6)by Sonazoid CEUS;76.5%(26/34)(95%CI:58.8-89.3) by Gd-EOB-DTPA MRI;and 88.2%(30/34)(95%CI: 72.5-96.7)by CT arterioportal angiography.The diagnostic sensitivity of detecting moderately-differentiated HCC by CECT,Sonazoid CEUS,Gd-EOB-DTPA MRI and CT arterioportal angiography was 62.5%(15/24)(95%CI: 40.6-81.2),79.2%(19/24)(95%CI:57.8-92.9),75.0% (18/24)(95%CI:53.3-90.2)and 95.8%(23/24)(95% CI:78.9-99.9),respectively.A significant difference(P< 0.05)was observed between CECT and CT arterioportal angiography in all nodules.There was no difference between Sonazoid CEUS,Gd-EOB-DTPA MRI,and CT arterioportal angiography.The combined sensitivity of Sonazoid CEUS and Gd-EOB-DTPA MRI was 94.1%(32/34).CONCLUSION:Changing the main diagnostic modality for HCC smaller than 2 cm from CT arterioportal angiography to Sonazoid CEUS and Gd-EOB-DTPA MRI is recommended.
基金The Marine S&T Fund of Shandong Province for Pilot National Laboratory for Marine Science and Technology(Qingdao)under contract Nos 2022QNLM050301-4 and 2021WHZZB1705the National Natural Science Foundation of China under contract Nos 41527901 and 42030406the National Key R&D Program of China under contract No.2019YFD0901001。
文摘Smaller mesoscale eddies(SMEs)have an important effect on the transmission of ocean temperatures,salinity,energy,and marine biochemical processes.However,traditional altimeters,the dominant sensors used to identify and track eddies,have made it challenging to observe SMEs accurately due to resolution limitations.Eddies drive local upwelling or downwelling,leaving signatures on sea surface temperatures(SSTs)and chlorophyll concentrations(Chls).SST can be observed by spaceborne infrared sensors,and Chl can be measured by ocean color remote sensing.Therefore,multisatellite observations provide an opportunity to obtain information to characterize SMEs.In this paper,an eddy detection algorithm based on SST and Chl images is proposed,which identifies eddies by characterizing the spatial and temporal distribution of SST and Chl data.The algorithm is applied to characterize and analyze SMEs in the Kuroshio Extension.Statistical results on their distribution and seasonal variability are shown,and the formation processes are preliminarily discussed.SMEs generation may be contributed by horizontal strain instability,the interaction of topographic obstacles and currents,and wind stress curl.
文摘China’s big spenders now far-flung by Yu Nan WIDELY recognized as one of the most beautiful cities in China,the coastal port of Dalian is a perfect example of a lowertier city becoming a growth hub for multinationals, owing to stronger purchasing power. That trend inspired Boston Consulting Group(BCG) to release a report,advising businessmen to look beyond the familiar mega-cities to the fast-growing smaller
基金The authors acknowledge the support from the National Key R&D Program of China(no.2022YFB3203600)the National Natural Science Foundation of China(no.52272162)+1 种基金the China Postdoctoral Science Foundation(Nos.2021T140535 and 2019M663696)the Alexander von Humboldt Foundation.L.T.thanks Dr.Christoph Meyer and Prof.Vasily Moshnyaga for their help in Raman spectroscopy measurement.M.L.acknowledges the support from Prof.Xixiang Zhang and the nanofabrication core lab at King Abdullah University of Science and Technology for the nanofabrication facilities.
文摘“Smaller is softer”is a reverse size dependence of strength,defying the“smaller is stronger”tenet.It usually results from surface-mediated displacive or diffusive deformation and is mainly found in some ultra-small-scale(below tens of nanometers)metallic materials.Here,making use of the surface modifi-cation via ion beam irradiation,we bring the“smaller is softer”into being in a covalently-bonded,hard,and brittle material-amorphous Si(a-Si)at a much larger size regime(<∼500 nm).It is manifested as the transition from the quasi-brittle failure to the homogeneous plastic deformation as well as the de-creasing yield stress with sample volume reduction at the submicron-scale regime.An analytical model of hard core/superplastic shell has been proposed to explain the artificially-controllable size-dependent softening.This surface engineering pathway via ion irradiation is not only of particular interest to tai-lor the strength and deformation behaviors in small-sized a-Si or other covalently-bonded amorphous solids but also of practical relevance to the utility of a-Si in microelectronics and microelectromechanical systems.
文摘Brain tumors are potentially fatal presence of cancer cells over a human brain,and they need to be segmented for accurate and reliable planning of diag-nosis.Segmentation process must be carried out in different regions based on which the stages of cancer can be accurately derived.Glioma patients exhibit a different level of challenge in terms of cancer or tumors detection as the Magnetic Resonance Imaging(MRI)images possess varying sizes,shapes,positions,and modalities.The scanner used for sensing the location of tumors cells will be sub-jected to additional protocols and measures for accuracy,in turn,increasing the time and affecting the performance of the entire model.In this view,Convolutional Neural Networks deliver suitable models for efficient segmentation and thus delivered promising results.The previous strategies and models failed to adhere to diversity of sizes and shapes,proving to be a well-established solution for detecting tumors of bigger size.Tumors tend to be smaller in size and shape during their premature stages and they can easily evade the algorithms of Convolutional Neural Network(CNN).This proposal intends to furnish a detailed model for sensing early stages of cancer and hence perform segmentation irrespective of the current size and shape of tumors.The size of networks and layers will lead to a significant weightage when multiple kernel sizes are involved,especially in multi-resolution environments.On the other hand,the proposed model is designed with a novel approach including a dilated convolution and level-based learning strat-egy.When the convolution process is dilated,the process of feature extraction deals with multiscale objective and level-based learning eliminates the shortcoming of previous models,thereby enhancing the quality of smaller tumors cells and shapes.The level-based learning approach also encapsulates the feature recon-struction processes which highlights the sensing of small-scale tumors growth.Inclusively,segmenting the images is performed with better accuracy and hence detection becomes better when compared to that of hierarchical approaches.
